[{"_1":2,"_448":-5,"_449":-5},"loaderData",{"_3":4,"_91":92,"_180":181,"_447":-5},"root",{"_5":6,"_32":33,"_60":61},"ctx",{"_7":8,"_19":20,"_21":-5,"_22":23,"_24":25,"_26":27,"_28":27,"_29":27,"_30":31},"auth",{"_9":-5,"_10":-5,"_11":-5,"_12":-5,"_13":-5,"_14":-5,"_15":-5,"_16":-5,"_17":-5,"_18":-5},"sessionClaims","sessionId","sessionStatus","userId","actor","orgId","orgRole","orgSlug","orgPermissions","factorVerificationAge","environmentVariableKeys",[],"user","preferencesKeybindings","default","preferencesTheme","light","preferencesFixers",false,"preferencesCopilot","preferencesSelectionAi","notificationCount",0,"ENV",{"_34":35,"_36":37,"_38":39,"_40":41,"_42":43,"_44":45,"_46":47,"_48":49,"_50":51,"_52":53,"_54":55,"_56":57,"_58":59},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","VAL_RUNTIME_NAME","prod","RENDER_GIT_COMMIT","d61e6fca584575ce46a18e0c1db610039aa1035f","RENDER_GIT_BRANCH","main","CLERK_PUBLISHABLE_KEY","pk_live_Y2xlcmsudmFsLnRvd24k","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","API_URL","https://api.val.town","WORKER_FINGERPRINT","fd7e47ba602df66e0f857b45e802b861","clerkState",{"_62":63},"__internal_clerk_state",{"_64":65,"_66":43,"_67":68,"_69":68,"_70":27,"_71":68,"_72":68,"_73":68,"_74":68,"_75":68,"_76":68,"_77":68,"_78":68,"_79":80,"_89":27,"_90":27},"__clerk_ssr_state",{"_9":-5,"_10":-5,"_11":-5,"_12":-5,"_13":-5,"_14":-5,"_15":-5,"_16":-5,"_17":-5,"_18":-5},"__publishableKey","__proxyUrl","","__domain","__isSatellite","__signInUrl","__signUpUrl","__afterSignInUrl","__afterSignUpUrl","__signInForceRedirectUrl","__signUpForceRedirectUrl","__signInFallbackRedirectUrl","__signUpFallbackRedirectUrl","__clerk_debug",{"_81":27,"_82":68,"_83":84,"_85":68,"_86":43,"_87":27,"_88":68},"isSignedIn","proxyUrl","reason","session-token-and-uat-missing","message","publishableKey","isSatellite","domain","__telemetryDisabled","__telemetryDebug","routes/_app.x.$handle.$name",{"_93":94,"_95":96,"_97":41,"_98":68,"_99":-7,"_100":101,"_138":139,"_161":162,"_163":164,"_171":172,"_176":-5,"_177":-5,"_178":27,"_179":27},"handle","mees","name","pinecone","branchName","path","version","project",{"_102":103,"_95":96,"_104":105,"_106":-5,"_107":27,"_108":109,"_110":-5,"_111":-5,"_12":112,"_113":-5,"_114":27,"_115":-5,"_116":31,"_117":118,"_119":118,"_120":121,"_122":121,"_123":121,"_124":118,"_125":27,"_21":126,"_129":130},"id","8ac31d58-3d3f-11f0-8942-76b3cceeab13","createdAt",["D",1748600356874],"deletedAt","legacyEnvVars","privacy","public","description","imageUrl","3e0d4fd0-3c52-11f0-82f4-07b6e52f9e9f","pinnedFileId","pinPreview","screenshotVersion","likeCount","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","isOwned",{"_93":94,"_127":128},"profileImageUrl","https://img.clerk.com/eyJ0eXBlIjoiZGVmYXVsdCIsImlpZCI6Imluc18yTm10cGhQUklwU0tqZHNiOHNnNGd2ZjJURFUiLCJyaWQiOiJ1c2VyXzJ4bDJlNkxSc2IxcmEyMG9KYkVyOVM4ekZmdyJ9","branch",{"_99":131,"_95":41,"_102":132,"_133":103,"_134":-5,"_135":-5,"_136":-5,"_137":-5},43,"8ac36ff6-3d3f-11f0-8942-76b3cceeab13","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectUserId","tree",[140],{"_102":141,"_95":142,"_143":144,"_145":-5,"_146":132,"_147":131,"_148":149,"_104":150,"_99":151,"_152":153,"_154":155,"_156":27,"_157":-5,"_158":159,"_98":160},"4a02f9f0-3e3e-11f0-b389-76b3cceeab13","new-file-6541.tsx","fileId","e35cf854-3d48-11f0-8002-76b3cceeab13","parentId","branchId","branchVersion","latest",true,["D",1748709769899],31,"fileType","http","content","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n// - PARSER_ASSISTANT_ID (optional)\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\nconst PARSER_ASSISTANT_ID = Deno.env.get(\"PARSER_ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Fast query parser using GPT-3.5\nasync function parseQueryIntent(query: string) {\n try {\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n\n // Create a quick thread for parsing\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({}),\n });\n\n const thread = await threadResponse.json();\n\n // Add the query\n await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n role: \"user\",\n content: query,\n }),\n });\n\n // Run the parser assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n assistant_id: PARSER_ASSISTANT_ID,\n max_completion_tokens: 200,\n }),\n });\n\n const run = await runResponse.json();\n\n // Poll for completion\n let runStatus = run;\n const maxWaitTime = 5000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Parser timeout\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 500));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n runStatus = await statusResponse.json();\n }\n\n // Get the response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=1`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n const messages = await messagesResponse.json();\n const assistantMessage = messages.data[0];\n\n if (assistantMessage && assistantMessage.content && assistantMessage.content[0]) {\n const content = assistantMessage.content[0].text.value;\n return JSON.parse(content);\n }\n\n throw new Error(\"No parser response\");\n } catch (error) {\n console.error(\"Parser error:\", error);\n // Return default if parser fails\n return {\n temporal: {\n tense: \"present\",\n explicit_date: null,\n recency_signal: \"none\",\n },\n explicit_filters: {\n company_only: null,\n exclude: null,\n },\n };\n }\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 50) {\n const index = pinecone.index(\"mees-articles\");\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(\"Filters:\", JSON.stringify(filters));\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build Pinecone filter from provided filters\n let pineconeFilter = {};\n\n // Only apply explicit filters\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone - let semantic search do the heavy lifting\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0);\n\n try {\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId, mediaFilters = {} } = await req.json();\n\n // Debug endpoint for testing specific queries - shows complete flow\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n\n let debugInfo = `🔍 COMPLETE QUERY FLOW DEBUG for: \"${testQuery}\"\\n${\"=\".repeat(80)}\\n\\n`;\n\n // Step 1: Parse the query\n debugInfo += `📝 STEP 1: QUERY PARSING\\n${\"-\".repeat(40)}\\n`;\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n try {\n parsedIntent = await parseQueryIntent(testQuery);\n debugInfo += `✅ Parser Results:\\n${JSON.stringify(parsedIntent, null, 2)}\\n\\n`;\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n debugInfo += `📊 Search Strategy: ${searchStrategy}\\n\\n`;\n } catch (e) {\n debugInfo += `❌ Parser Error: ${e.message}\\n\\n`;\n }\n } else {\n debugInfo += `⚠️ Parser not configured (PARSER_ASSISTANT_ID missing)\\n\\n`;\n }\n\n // Step 2: Build filters\n debugInfo += `🔧 STEP 2: FILTER BUILDING\\n${\"-\".repeat(40)}\\n`;\n const baseFilters = {};\n if (mediaFilters?.requireCharts) baseFilters.has_charts = true;\n if (mediaFilters?.requireMaps) baseFilters.has_maps = true;\n if (mediaFilters?.requireTables) baseFilters.has_tables = true;\n\n // Apply explicit company filter if specified\n if (parsedIntent?.explicit_filters?.company_only) {\n baseFilters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n debugInfo += `📋 Final Filters for Pinecone:\\n${JSON.stringify(baseFilters, null, 2)}\\n\\n`;\n\n // Step 3: Run Pinecone search\n debugInfo += `🔎 STEP 3: PINECONE SEARCH\\n${\"-\".repeat(40)}\\n`;\n try {\n const searchResults = await searchPinecone(testQuery, baseFilters, 20);\n\n debugInfo += `✅ Found ${searchResults.length} results\\n\\n`;\n debugInfo += `📊 Top 5 Results Summary:\\n`;\n\n searchResults.slice(0, 5).forEach((result, i) => {\n const meta = result.metadata;\n debugInfo += `\\n[${i + 1}] Score: ${result.score.toFixed(4)}\\n`;\n debugInfo += ` Title: ${meta.title ? meta.title.substring(0, 80) : \"No title\"}\\n`;\n debugInfo += ` Date: ${meta.publish_date} | Article ID: ${meta.article_id}\\n`;\n debugInfo += ` Charts: ${meta.all_charts || \"[]\"}\\n`;\n debugInfo += ` Maps: ${meta.all_maps || \"[]\"}\\n`;\n debugInfo += ` Tables: ${meta.all_tables || \"[]\"}\\n`;\n debugInfo += ` Companies: ${meta.companies || \"[]\"}\\n`;\n });\n\n // Step 4: Format data for OpenAI\n debugInfo += `\\n\\n📦 STEP 4: DATA FORMATTING FOR OPENAI\\n${\"-\".repeat(40)}\\n`;\n\n const formattedResults = searchResults.slice(0, 10).map((match) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Truncate text for OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500;\n if (textContent.length > maxTextLength) {\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n };\n });\n\n // Sort by date\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n debugInfo += `✅ Formatted ${formattedResults.length} results for OpenAI\\n`;\n debugInfo += `📄 Sample formatted result:\\n${\n JSON.stringify(formattedResults[0], null, 2).substring(0, 800)\n }...\\n\\n`;\n\n // Step 5: Prepare message for assistant\n debugInfo += `💬 STEP 5: MESSAGE PREPARATION FOR ASSISTANT\\n${\"-\".repeat(40)}\\n`;\n\n let enhancedMessage = testQuery;\n if (parsedIntent?.temporal?.tense === \"present\") {\n enhancedMessage +=\n \"\\n\\n[System Note: User query is present-tense. Prioritize recent data (marked as PRIMARY_SOURCE) for current facts and figures. Use older articles (marked as HISTORICAL_CONTEXT) only for background and trends.]\";\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from search results are automatically displayed to the user in a side panel.]\";\n }\n\n debugInfo += `📝 Enhanced Message:\\n\"${enhancedMessage}\"\\n\\n`;\n\n // Step 6: Simulate what would be sent to assistant\n debugInfo += `🤖 STEP 6: ASSISTANT INTERACTION SUMMARY\\n${\"-\".repeat(40)}\\n`;\n debugInfo += `Assistant ID: ${ASSISTANT_ID}\\n`;\n debugInfo += `Would create/use thread and send:\\n`;\n debugInfo += `1. User message: \"${testQuery}\"\\n`;\n debugInfo += `2. Tool calls would include:\\n`;\n debugInfo += ` - search_articles(query=\"${testQuery}\", filters=${JSON.stringify(baseFilters)})\\n`;\n debugInfo += `3. Tool would return ${formattedResults.length} formatted results\\n`;\n debugInfo += `4. Assistant would process and generate response\\n\\n`;\n\n // Media auto-display summary\n debugInfo += `🖼️ MEDIA AUTO-DISPLAY\\n${\"-\".repeat(40)}\\n`;\n const autoDisplayMedia = [];\n formattedResults.slice(0, 5).forEach((result) => {\n result.all_charts.forEach(id => autoDisplayMedia.push({ id, type: \"chart\", from: result.title }));\n result.all_maps.forEach(id => autoDisplayMedia.push({ id, type: \"map\", from: result.title }));\n result.all_tables.slice(0, 2).forEach(id => autoDisplayMedia.push({ id, type: \"table\", from: result.title }));\n });\n\n debugInfo += `Would auto-display ${autoDisplayMedia.length} media items:\\n`;\n autoDisplayMedia.slice(0, 10).forEach(m => {\n debugInfo += ` - ${m.type} ${m.id} from \"${m.from.substring(0, 50)}...\"\\n`;\n });\n } catch (searchError) {\n debugInfo += `❌ Search Error: ${searchError.message}\\n`;\n }\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n has_charts: meta.has_charts,\n has_maps: meta.has_maps,\n has_tables: meta.has_tables,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) =>\n new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime()\n );\n\n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n\n // Format the response with all metadata\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10 with full metadata:\\n`;\n\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n articleInfo += ` 🏷️ Has Media: Charts=${d.has_charts} Maps=${d.has_maps} Tables=${d.has_tables}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === \"string\") {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || \"N/A\"}: ${parsed.context || \"N/A\"}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === \"object\") {\n // If it's already an object\n articleInfo += ` • ${km.metric || \"N/A\"}: ${km.context || \"N/A\"}\\n`;\n }\n });\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return new Response(\n JSON.stringify({\n response: summary + formattedArticles,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Parse the query first (if parser assistant is configured)\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n console.log(\"Parsing query with fast assistant...\");\n parsedIntent = await parseQueryIntent(message);\n console.log(\"Parsed intent:\", parsedIntent);\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n }\n\n console.log(`Search strategy: ${searchStrategy}`);\n\n // Build base filters from UI toggles\n let baseFilters = {};\n if (mediaFilters.requireCharts) baseFilters.has_charts = true;\n if (mediaFilters.requireMaps) baseFilters.has_maps = true;\n if (mediaFilters.requireTables) baseFilters.has_tables = true;\n\n // Apply explicit company filter if specified\n if (parsedIntent?.explicit_filters?.company_only) {\n baseFilters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n\n // Add context based on parsed intent\n if (parsedIntent?.temporal?.tense === \"present\") {\n enhancedMessage +=\n \"\\n\\n[System Note: User query is present-tense. Prioritize recent data (marked as PRIMARY_SOURCE) for current facts and figures. Use older articles (marked as HISTORICAL_CONTEXT) only for background and trends.]\";\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from search results are automatically displayed to the user in a side panel.]\";\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[], context: any) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n\n // Use the context passed to this function\n const searchQuery = parsedArgs.query;\n const searchFilters = { ...context.baseFilters, ...(parsedArgs.filters || {}) };\n\n // Get more results initially\n const results = await searchPinecone(\n searchQuery,\n searchFilters,\n 50, // Get more results for filtering\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format and categorize results based on search strategy\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500;\n if (textContent.length > maxTextLength) {\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // Categorize results based on search strategy\n let finalResults = formattedResults;\n\n if (context.searchStrategy === \"recent_focused\") {\n // For present-tense queries, prioritize recent data\n const sixMonthsAgo = new Date();\n sixMonthsAgo.setMonth(sixMonthsAgo.getMonth() - 6);\n\n const recentResults = formattedResults.filter(r => new Date(r.publish_date_iso || r.date) > sixMonthsAgo);\n\n const historicalResults = formattedResults.filter(r =>\n new Date(r.publish_date_iso || r.date) <= sixMonthsAgo\n ).slice(0, 5);\n\n // Mark results with their context\n recentResults.forEach(r => r._context = \"PRIMARY_SOURCE\");\n historicalResults.forEach(r => r._context = \"HISTORICAL_CONTEXT\");\n\n // Combine with recent first\n finalResults = [...recentResults.slice(0, 20), ...historicalResults];\n }\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n\n // Use top results for media display (prioritize PRIMARY_SOURCE)\n let resultsToScan = finalResults\n .filter(r => !r._context || r._context === \"PRIMARY_SOURCE\")\n .slice(0, 10);\n\n if (resultsToScan.length < 5) {\n resultsToScan = finalResults.slice(0, 10);\n }\n\n resultsToScan.forEach((result, index) => {\n // Add all media types from results\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(\n \"Results after sorting:\",\n finalResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n context: r._context || \"STANDARD\",\n })),\n );\n\n console.log(`Returning ${finalResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(finalResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders },\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n // Pass context to handleToolCalls\n const context = {\n baseFilters,\n searchStrategy,\n parsedIntent,\n };\n\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls, context);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders },\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}","deleted","label","children",[],[142],"files",[140],"counts",{"_165":31,"_166":167,"_168":31,"_169":170},"pulls","branches",1,"forks","envVars",4,"contents",[173],{"_102":141,"_95":142,"_143":144,"_145":-5,"_146":132,"_147":131,"_148":149,"_104":174,"_99":151,"_152":153,"_154":155,"_156":27,"_157":-5,"_98":175},["D",1748709769899],[142],"file","upstream","emptyEnvVars","isNew","routes/_app.x.$handle.$name.branch.$branch.history",{"_93":94,"_95":96,"_97":41,"_100":182,"_186":187,"_445":31,"_446":149},{"_102":103,"_95":96,"_104":183,"_106":-5,"_107":27,"_108":109,"_110":-5,"_111":-5,"_12":112,"_113":-5,"_114":27,"_115":-5,"_116":31,"_117":118,"_119":118,"_120":121,"_122":121,"_123":121,"_124":118,"_125":27,"_21":184,"_129":185},["D",1748600356874],{"_93":94,"_127":128},{"_99":131,"_95":41,"_102":132,"_133":103,"_134":-5,"_135":-5,"_136":-5,"_137":-5},"history",[188,193,202,211,220,228,237,245,254,263,272,281,289,297,305,313,321,329,337,345,353,361,369,377,385,393,401,409,416,424,432,438],{"_143":144,"_99":131,"_104":189,"_146":132,"_161":190},["D",1748709769899],[191],{"_102":141,"_146":132,"_143":144,"_95":142,"_154":155,"_152":153,"_99":151,"_145":-5,"_104":189,"_147":131,"_157":-5,"_148":149,"_156":27,"_98":192},[142],{"_143":144,"_99":194,"_104":195,"_146":132,"_161":196},42,["D",1748709653025],[197],{"_102":198,"_146":132,"_143":144,"_95":142,"_154":199,"_152":153,"_99":200,"_145":-5,"_104":195,"_147":194,"_157":-5,"_148":27,"_156":27,"_98":201},"04598bc6-3e3e-11f0-bbcb-76b3cceeab13","// MEES Assistant API Endpoint - Cleaned Version\n// Required environment variables:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n// - PARSER_ASSISTANT_ID (optional)\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Initialize clients\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\nconst PARSER_ASSISTANT_ID = Deno.env.get(\"PARSER_ASSISTANT_ID\");\n\n// Media ID to Title mapping\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Fast query parser using GPT-3.5\nasync function parseQueryIntent(query: string) {\n try {\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n\n // Create a quick thread for parsing\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({}),\n });\n\n const thread = await threadResponse.json();\n\n // Add the query\n await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n role: \"user\",\n content: query,\n }),\n });\n\n // Run the parser assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n assistant_id: PARSER_ASSISTANT_ID,\n max_completion_tokens: 200,\n }),\n });\n\n const run = await runResponse.json();\n\n // Poll for completion\n let runStatus = run;\n const maxWaitTime = 5000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Parser timeout\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 500));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n runStatus = await statusResponse.json();\n }\n\n // Get the response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=1`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n const messages = await messagesResponse.json();\n const assistantMessage = messages.data[0];\n\n if (assistantMessage && assistantMessage.content && assistantMessage.content[0]) {\n const content = assistantMessage.content[0].text.value;\n return JSON.parse(content);\n }\n\n throw new Error(\"No parser response\");\n } catch (error) {\n console.error(\"Parser error:\", error);\n // Return default if parser fails\n return {\n temporal: {\n tense: \"present\",\n explicit_date: null,\n recency_signal: \"none\",\n },\n explicit_filters: {\n company_only: null,\n exclude: null,\n },\n };\n }\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 50) {\n const index = pinecone.index(\"mees-articles\");\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(filters).length > 0 ? filters : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n return searchResults.matches;\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces)\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces)\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\n// Debug endpoint for testing queries\nasync function handleDebugQuery(testQuery: string) {\n let debugInfo = `🔍 COMPLETE QUERY FLOW DEBUG for: \"${testQuery}\"\\n${\"=\".repeat(80)}\\n\\n`;\n\n // Step 1: Parse the query\n debugInfo += `📝 STEP 1: QUERY PARSING\\n${\"-\".repeat(40)}\\n`;\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n try {\n parsedIntent = await parseQueryIntent(testQuery);\n debugInfo += `✅ Parser Results:\\n${JSON.stringify(parsedIntent, null, 2)}\\n\\n`;\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n debugInfo += `📊 Search Strategy: ${searchStrategy}\\n\\n`;\n } catch (e) {\n debugInfo += `❌ Parser Error: ${e.message}\\n\\n`;\n }\n } else {\n debugInfo += `⚠️ Parser not configured (PARSER_ASSISTANT_ID missing)\\n\\n`;\n }\n\n // Step 2: Build filters\n debugInfo += `🔧 STEP 2: FILTER BUILDING\\n${\"-\".repeat(40)}\\n`;\n const filters = {};\n\n // Apply explicit company filter if specified\n if (parsedIntent?.explicit_filters?.company_only) {\n filters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n debugInfo += `📋 Final Filters for Pinecone:\\n${JSON.stringify(filters, null, 2)}\\n\\n`;\n\n // Step 3: Run Pinecone search\n debugInfo += `🔎 STEP 3: PINECONE SEARCH\\n${\"-\".repeat(40)}\\n`;\n try {\n const searchResults = await searchPinecone(testQuery, filters, 20);\n\n debugInfo += `✅ Found ${searchResults.length} results\\n\\n`;\n debugInfo += `📊 Top 5 Results Summary:\\n`;\n\n searchResults.slice(0, 5).forEach((result, i) => {\n const meta = result.metadata;\n debugInfo += `\\n[${i + 1}] Score: ${result.score.toFixed(4)}\\n`;\n debugInfo += ` Title: ${meta.title ? meta.title.substring(0, 80) : \"No title\"}\\n`;\n debugInfo += ` Date: ${meta.publish_date} | Article ID: ${meta.article_id}\\n`;\n debugInfo += ` Charts: ${meta.all_charts || \"[]\"}\\n`;\n debugInfo += ` Maps: ${meta.all_maps || \"[]\"}\\n`;\n debugInfo += ` Tables: ${meta.all_tables || \"[]\"}\\n`;\n debugInfo += ` Companies: ${meta.companies || \"[]\"}\\n`;\n });\n\n // Step 4: Show what would be sent to assistant\n debugInfo += `\\n\\n🤖 STEP 4: ASSISTANT INTERACTION SUMMARY\\n${\"-\".repeat(40)}\\n`;\n debugInfo += `Assistant ID: ${ASSISTANT_ID}\\n`;\n debugInfo += `Would create/use thread and send:\\n`;\n debugInfo += `1. User message: \"${testQuery}\"\\n`;\n debugInfo += `2. Tool calls would include:\\n`;\n debugInfo += ` - search_articles(query=\"${testQuery}\", filters=${JSON.stringify(filters)})\\n`;\n debugInfo += `3. Tool would return ${searchResults.length} formatted results\\n`;\n debugInfo += `4. Assistant would process and generate response\\n\\n`;\n\n // Media auto-display summary\n debugInfo += `🖼️ MEDIA AUTO-DISPLAY\\n${\"-\".repeat(40)}\\n`;\n const autoDisplayMedia = [];\n searchResults.slice(0, 5).forEach((result) => {\n const meta = result.metadata;\n const charts = JSON.parse(meta.all_charts || \"[]\");\n const maps = JSON.parse(meta.all_maps || \"[]\");\n const tables = JSON.parse(meta.all_tables || \"[]\");\n\n charts.forEach(id => autoDisplayMedia.push({ id, type: \"chart\", from: meta.title }));\n maps.forEach(id => autoDisplayMedia.push({ id, type: \"map\", from: meta.title }));\n tables.slice(0, 2).forEach(id => autoDisplayMedia.push({ id, type: \"table\", from: meta.title }));\n });\n\n debugInfo += `Would auto-display ${autoDisplayMedia.length} media items:\\n`;\n autoDisplayMedia.slice(0, 10).forEach(m => {\n debugInfo += ` - ${m.type} ${m.id} from \"${m.from.substring(0, 50)}...\"\\n`;\n });\n } catch (searchError) {\n debugInfo += `❌ Search Error: ${searchError.message}\\n`;\n }\n\n return debugInfo;\n}\n\n// Debug endpoint for testing dates\nasync function handleDebugDates() {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) => new Date(b.date_iso || b.date).getTime() - new Date(a.date_iso || a.date).getTime());\n\n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n\n // Format the response\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10:\\n`;\n\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return summary + formattedArticles;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoints\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n const debugInfo = await handleDebugQuery(testQuery);\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n if (message === \"DEBUG_DATES\") {\n const debugInfo = await handleDebugDates();\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n // Verify required environment variables\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // OpenAI API headers\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n };\n\n console.log(\"Processing message:\", message.substring(0, 100) + \"...\");\n\n // Parse the query first (if parser assistant is configured)\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n console.log(\"Parsing query with fast assistant...\");\n parsedIntent = await parseQueryIntent(message);\n console.log(\"Parsed intent:\", parsedIntent);\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n }\n\n console.log(`Search strategy: ${searchStrategy}`);\n\n // Build filters from parsed intent\n let filters = {};\n if (parsedIntent?.explicit_filters?.company_only) {\n filters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n // Create or retrieve thread\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n\n // Add context based on parsed intent\n if (parsedIntent?.temporal?.tense === \"present\") {\n enhancedMessage +=\n \"\\n\\n[System Note: User query is present-tense. Prioritize recent data (marked as PRIMARY_SOURCE) for current facts and figures. Use older articles (marked as HISTORICAL_CONTEXT) only for background and trends.]\";\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from search results are automatically displayed to the user in a side panel.]\";\n }\n\n // Add message to thread\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[], context: any) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n console.log(`Raw arguments:`, args);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Parsed arguments:`, JSON.stringify(parsedArgs));\n console.log(`Arguments:`, JSON.stringify(parsedArgs));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n\n // Use the context passed to this function\n const searchQuery = parsedArgs.query || context.originalMessage;\n const searchFilters = { ...context.baseFilters, ...(parsedArgs.filters || {}) };\n\n if (!searchQuery) {\n throw new Error(\"No search query provided\");\n }\n\n console.log(`Actual search query: \"${searchQuery}\"`);\n\n // Get results\n const results = await searchPinecone(searchQuery, searchFilters, 50);\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Truncate text for OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500;\n if (textContent.length > maxTextLength) {\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date - newest first\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // Apply search strategy\n let finalResults = formattedResults;\n\n if (context.searchStrategy === \"recent_focused\") {\n // For present-tense queries, prioritize recent data\n const sixMonthsAgo = new Date();\n sixMonthsAgo.setMonth(sixMonthsAgo.getMonth() - 6);\n\n const recentResults = formattedResults.filter(r => new Date(r.publish_date_iso || r.date) > sixMonthsAgo);\n const historicalResults = formattedResults.filter(r =>\n new Date(r.publish_date_iso || r.date) <= sixMonthsAgo\n ).slice(0, 5);\n\n // Mark results with their context\n recentResults.forEach(r => r._context = \"PRIMARY_SOURCE\");\n historicalResults.forEach(r => r._context = \"HISTORICAL_CONTEXT\");\n\n // Combine with recent first\n finalResults = [...recentResults.slice(0, 20), ...historicalResults];\n }\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n\n // Use top results for media display\n let resultsToScan = finalResults\n .filter(r => !r._context || r._context === \"PRIMARY_SOURCE\")\n .slice(0, 10);\n\n if (resultsToScan.length < 5) {\n resultsToScan = finalResults.slice(0, 10);\n }\n\n resultsToScan.forEach((result) => {\n // Add all media types from results\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(`Returning ${finalResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(finalResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders },\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n // Pass context to handleToolCalls\n const context = {\n baseFilters: filters,\n searchStrategy,\n parsedIntent,\n originalMessage: message,\n };\n\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls, context);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders },\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",30,[142],{"_143":144,"_99":203,"_104":204,"_146":132,"_161":205},41,["D",1748709492382],[206],{"_102":207,"_146":132,"_143":144,"_95":142,"_154":208,"_152":153,"_99":209,"_145":-5,"_104":204,"_147":203,"_157":-5,"_148":27,"_156":27,"_98":210},"a4990860-3e3d-11f0-81c4-76b3cceeab13","// MEES Assistant API Endpoint - Cleaned Version\n// Required environment variables:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n// - PARSER_ASSISTANT_ID (optional)\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Initialize clients\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\nconst PARSER_ASSISTANT_ID = Deno.env.get(\"PARSER_ASSISTANT_ID\");\n\n// Media ID to Title mapping\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Fast query parser using GPT-3.5\nasync function parseQueryIntent(query: string) {\n try {\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n\n // Create a quick thread for parsing\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({}),\n });\n\n const thread = await threadResponse.json();\n\n // Add the query\n await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n role: \"user\",\n content: query,\n }),\n });\n\n // Run the parser assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n assistant_id: PARSER_ASSISTANT_ID,\n max_completion_tokens: 200,\n }),\n });\n\n const run = await runResponse.json();\n\n // Poll for completion\n let runStatus = run;\n const maxWaitTime = 5000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Parser timeout\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 500));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n runStatus = await statusResponse.json();\n }\n\n // Get the response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=1`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n const messages = await messagesResponse.json();\n const assistantMessage = messages.data[0];\n\n if (assistantMessage && assistantMessage.content && assistantMessage.content[0]) {\n const content = assistantMessage.content[0].text.value;\n return JSON.parse(content);\n }\n\n throw new Error(\"No parser response\");\n } catch (error) {\n console.error(\"Parser error:\", error);\n // Return default if parser fails\n return {\n temporal: {\n tense: \"present\",\n explicit_date: null,\n recency_signal: \"none\",\n },\n explicit_filters: {\n company_only: null,\n exclude: null,\n },\n };\n }\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 50) {\n const index = pinecone.index(\"mees-articles\");\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(filters).length > 0 ? filters : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n return searchResults.matches;\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces)\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces)\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\n// Debug endpoint for testing queries\nasync function handleDebugQuery(testQuery: string) {\n let debugInfo = `🔍 COMPLETE QUERY FLOW DEBUG for: \"${testQuery}\"\\n${\"=\".repeat(80)}\\n\\n`;\n\n // Step 1: Parse the query\n debugInfo += `📝 STEP 1: QUERY PARSING\\n${\"-\".repeat(40)}\\n`;\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n try {\n parsedIntent = await parseQueryIntent(testQuery);\n debugInfo += `✅ Parser Results:\\n${JSON.stringify(parsedIntent, null, 2)}\\n\\n`;\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n debugInfo += `📊 Search Strategy: ${searchStrategy}\\n\\n`;\n } catch (e) {\n debugInfo += `❌ Parser Error: ${e.message}\\n\\n`;\n }\n } else {\n debugInfo += `⚠️ Parser not configured (PARSER_ASSISTANT_ID missing)\\n\\n`;\n }\n\n // Step 2: Build filters\n debugInfo += `🔧 STEP 2: FILTER BUILDING\\n${\"-\".repeat(40)}\\n`;\n const filters = {};\n\n // Apply explicit company filter if specified\n if (parsedIntent?.explicit_filters?.company_only) {\n filters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n debugInfo += `📋 Final Filters for Pinecone:\\n${JSON.stringify(filters, null, 2)}\\n\\n`;\n\n // Step 3: Run Pinecone search\n debugInfo += `🔎 STEP 3: PINECONE SEARCH\\n${\"-\".repeat(40)}\\n`;\n try {\n const searchResults = await searchPinecone(testQuery, filters, 20);\n\n debugInfo += `✅ Found ${searchResults.length} results\\n\\n`;\n debugInfo += `📊 Top 5 Results Summary:\\n`;\n\n searchResults.slice(0, 5).forEach((result, i) => {\n const meta = result.metadata;\n debugInfo += `\\n[${i + 1}] Score: ${result.score.toFixed(4)}\\n`;\n debugInfo += ` Title: ${meta.title ? meta.title.substring(0, 80) : \"No title\"}\\n`;\n debugInfo += ` Date: ${meta.publish_date} | Article ID: ${meta.article_id}\\n`;\n debugInfo += ` Charts: ${meta.all_charts || \"[]\"}\\n`;\n debugInfo += ` Maps: ${meta.all_maps || \"[]\"}\\n`;\n debugInfo += ` Tables: ${meta.all_tables || \"[]\"}\\n`;\n debugInfo += ` Companies: ${meta.companies || \"[]\"}\\n`;\n });\n\n // Step 4: Show what would be sent to assistant\n debugInfo += `\\n\\n🤖 STEP 4: ASSISTANT INTERACTION SUMMARY\\n${\"-\".repeat(40)}\\n`;\n debugInfo += `Assistant ID: ${ASSISTANT_ID}\\n`;\n debugInfo += `Would create/use thread and send:\\n`;\n debugInfo += `1. User message: \"${testQuery}\"\\n`;\n debugInfo += `2. Tool calls would include:\\n`;\n debugInfo += ` - search_articles(query=\"${testQuery}\", filters=${JSON.stringify(filters)})\\n`;\n debugInfo += `3. Tool would return ${searchResults.length} formatted results\\n`;\n debugInfo += `4. Assistant would process and generate response\\n\\n`;\n\n // Media auto-display summary\n debugInfo += `🖼️ MEDIA AUTO-DISPLAY\\n${\"-\".repeat(40)}\\n`;\n const autoDisplayMedia = [];\n searchResults.slice(0, 5).forEach((result) => {\n const meta = result.metadata;\n const charts = JSON.parse(meta.all_charts || \"[]\");\n const maps = JSON.parse(meta.all_maps || \"[]\");\n const tables = JSON.parse(meta.all_tables || \"[]\");\n\n charts.forEach(id => autoDisplayMedia.push({ id, type: \"chart\", from: meta.title }));\n maps.forEach(id => autoDisplayMedia.push({ id, type: \"map\", from: meta.title }));\n tables.slice(0, 2).forEach(id => autoDisplayMedia.push({ id, type: \"table\", from: meta.title }));\n });\n\n debugInfo += `Would auto-display ${autoDisplayMedia.length} media items:\\n`;\n autoDisplayMedia.slice(0, 10).forEach(m => {\n debugInfo += ` - ${m.type} ${m.id} from \"${m.from.substring(0, 50)}...\"\\n`;\n });\n } catch (searchError) {\n debugInfo += `❌ Search Error: ${searchError.message}\\n`;\n }\n\n return debugInfo;\n}\n\n// Debug endpoint for testing dates\nasync function handleDebugDates() {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) => new Date(b.date_iso || b.date).getTime() - new Date(a.date_iso || a.date).getTime());\n\n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n\n // Format the response\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10:\\n`;\n\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return summary + formattedArticles;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoints\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n const debugInfo = await handleDebugQuery(testQuery);\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n if (message === \"DEBUG_DATES\") {\n const debugInfo = await handleDebugDates();\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n // Verify required environment variables\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // OpenAI API headers\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n };\n\n console.log(\"Processing message:\", message.substring(0, 100) + \"...\");\n\n // Parse the query first (if parser assistant is configured)\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n console.log(\"Parsing query with fast assistant...\");\n parsedIntent = await parseQueryIntent(message);\n console.log(\"Parsed intent:\", parsedIntent);\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n }\n\n console.log(`Search strategy: ${searchStrategy}`);\n\n // Build filters from parsed intent\n let filters = {};\n if (parsedIntent?.explicit_filters?.company_only) {\n filters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n // Create or retrieve thread\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n\n // Add context based on parsed intent\n if (parsedIntent?.temporal?.tense === \"present\") {\n enhancedMessage +=\n \"\\n\\n[System Note: User query is present-tense. Prioritize recent data (marked as PRIMARY_SOURCE) for current facts and figures. Use older articles (marked as HISTORICAL_CONTEXT) only for background and trends.]\";\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from search results are automatically displayed to the user in a side panel.]\";\n }\n\n // Add message to thread\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[], context: any) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n\n // Use the context passed to this function\n const searchQuery = parsedArgs.query;\n const searchFilters = { ...context.baseFilters, ...(parsedArgs.filters || {}) };\n\n // Get results\n const results = await searchPinecone(searchQuery, searchFilters, 50);\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Truncate text for OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500;\n if (textContent.length > maxTextLength) {\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date - newest first\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // Apply search strategy\n let finalResults = formattedResults;\n\n if (context.searchStrategy === \"recent_focused\") {\n // For present-tense queries, prioritize recent data\n const sixMonthsAgo = new Date();\n sixMonthsAgo.setMonth(sixMonthsAgo.getMonth() - 6);\n\n const recentResults = formattedResults.filter(r => new Date(r.publish_date_iso || r.date) > sixMonthsAgo);\n const historicalResults = formattedResults.filter(r =>\n new Date(r.publish_date_iso || r.date) <= sixMonthsAgo\n ).slice(0, 5);\n\n // Mark results with their context\n recentResults.forEach(r => r._context = \"PRIMARY_SOURCE\");\n historicalResults.forEach(r => r._context = \"HISTORICAL_CONTEXT\");\n\n // Combine with recent first\n finalResults = [...recentResults.slice(0, 20), ...historicalResults];\n }\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n\n // Use top results for media display\n let resultsToScan = finalResults\n .filter(r => !r._context || r._context === \"PRIMARY_SOURCE\")\n .slice(0, 10);\n\n if (resultsToScan.length < 5) {\n resultsToScan = finalResults.slice(0, 10);\n }\n\n resultsToScan.forEach((result) => {\n // Add all media types from results\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(`Returning ${finalResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(finalResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders },\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n // Pass context to handleToolCalls\n const context = {\n baseFilters: filters,\n searchStrategy,\n parsedIntent,\n };\n\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls, context);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders },\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",29,[142],{"_143":144,"_99":212,"_104":213,"_146":132,"_161":214},40,["D",1748709259248],[215],{"_102":216,"_146":132,"_143":144,"_95":142,"_154":217,"_152":153,"_99":218,"_145":-5,"_104":213,"_147":212,"_157":-5,"_148":27,"_156":27,"_98":219},"19a4562e-3e3d-11f0-a0c1-76b3cceeab13","// MEES Assistant API Endpoint - Cleaned Version\n// Required environment variables:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n// - PARSER_ASSISTANT_ID (optional)\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Initialize clients\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\nconst PARSER_ASSISTANT_ID = Deno.env.get(\"PARSER_ASSISTANT_ID\");\n\n// Media ID to Title mapping\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Fast query parser using GPT-3.5\nasync function parseQueryIntent(query: string) {\n try {\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n\n // Create a quick thread for parsing\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({}),\n });\n\n const thread = await threadResponse.json();\n\n // Add the query\n await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n role: \"user\",\n content: query,\n }),\n });\n\n // Run the parser assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n assistant_id: PARSER_ASSISTANT_ID,\n max_completion_tokens: 200,\n }),\n });\n\n const run = await runResponse.json();\n\n // Poll for completion\n let runStatus = run;\n const maxWaitTime = 5000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Parser timeout\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 500));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n runStatus = await statusResponse.json();\n }\n\n // Get the response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=1`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n const messages = await messagesResponse.json();\n const assistantMessage = messages.data[0];\n\n if (assistantMessage && assistantMessage.content && assistantMessage.content[0]) {\n const content = assistantMessage.content[0].text.value;\n return JSON.parse(content);\n }\n\n throw new Error(\"No parser response\");\n } catch (error) {\n console.error(\"Parser error:\", error);\n // Return default if parser fails\n return {\n temporal: {\n tense: \"present\",\n explicit_date: null,\n recency_signal: \"none\",\n },\n explicit_filters: {\n company_only: null,\n exclude: null,\n },\n };\n }\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 50) {\n const index = pinecone.index(\"mees-articles\");\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: filters,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n return searchResults.matches;\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces)\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces)\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\n// Debug endpoint for testing queries\nasync function handleDebugQuery(testQuery: string) {\n let debugInfo = `🔍 COMPLETE QUERY FLOW DEBUG for: \"${testQuery}\"\\n${\"=\".repeat(80)}\\n\\n`;\n\n // Step 1: Parse the query\n debugInfo += `📝 STEP 1: QUERY PARSING\\n${\"-\".repeat(40)}\\n`;\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n try {\n parsedIntent = await parseQueryIntent(testQuery);\n debugInfo += `✅ Parser Results:\\n${JSON.stringify(parsedIntent, null, 2)}\\n\\n`;\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n debugInfo += `📊 Search Strategy: ${searchStrategy}\\n\\n`;\n } catch (e) {\n debugInfo += `❌ Parser Error: ${e.message}\\n\\n`;\n }\n } else {\n debugInfo += `⚠️ Parser not configured (PARSER_ASSISTANT_ID missing)\\n\\n`;\n }\n\n // Step 2: Build filters\n debugInfo += `🔧 STEP 2: FILTER BUILDING\\n${\"-\".repeat(40)}\\n`;\n const filters = {};\n\n // Apply explicit company filter if specified\n if (parsedIntent?.explicit_filters?.company_only) {\n filters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n debugInfo += `📋 Final Filters for Pinecone:\\n${JSON.stringify(filters, null, 2)}\\n\\n`;\n\n // Step 3: Run Pinecone search\n debugInfo += `🔎 STEP 3: PINECONE SEARCH\\n${\"-\".repeat(40)}\\n`;\n try {\n const searchResults = await searchPinecone(testQuery, filters, 20);\n\n debugInfo += `✅ Found ${searchResults.length} results\\n\\n`;\n debugInfo += `📊 Top 5 Results Summary:\\n`;\n\n searchResults.slice(0, 5).forEach((result, i) => {\n const meta = result.metadata;\n debugInfo += `\\n[${i + 1}] Score: ${result.score.toFixed(4)}\\n`;\n debugInfo += ` Title: ${meta.title ? meta.title.substring(0, 80) : \"No title\"}\\n`;\n debugInfo += ` Date: ${meta.publish_date} | Article ID: ${meta.article_id}\\n`;\n debugInfo += ` Charts: ${meta.all_charts || \"[]\"}\\n`;\n debugInfo += ` Maps: ${meta.all_maps || \"[]\"}\\n`;\n debugInfo += ` Tables: ${meta.all_tables || \"[]\"}\\n`;\n debugInfo += ` Companies: ${meta.companies || \"[]\"}\\n`;\n });\n\n // Step 4: Show what would be sent to assistant\n debugInfo += `\\n\\n🤖 STEP 4: ASSISTANT INTERACTION SUMMARY\\n${\"-\".repeat(40)}\\n`;\n debugInfo += `Assistant ID: ${ASSISTANT_ID}\\n`;\n debugInfo += `Would create/use thread and send:\\n`;\n debugInfo += `1. User message: \"${testQuery}\"\\n`;\n debugInfo += `2. Tool calls would include:\\n`;\n debugInfo += ` - search_articles(query=\"${testQuery}\", filters=${JSON.stringify(filters)})\\n`;\n debugInfo += `3. Tool would return ${searchResults.length} formatted results\\n`;\n debugInfo += `4. Assistant would process and generate response\\n\\n`;\n\n // Media auto-display summary\n debugInfo += `🖼️ MEDIA AUTO-DISPLAY\\n${\"-\".repeat(40)}\\n`;\n const autoDisplayMedia = [];\n searchResults.slice(0, 5).forEach((result) => {\n const meta = result.metadata;\n const charts = JSON.parse(meta.all_charts || \"[]\");\n const maps = JSON.parse(meta.all_maps || \"[]\");\n const tables = JSON.parse(meta.all_tables || \"[]\");\n\n charts.forEach(id => autoDisplayMedia.push({ id, type: \"chart\", from: meta.title }));\n maps.forEach(id => autoDisplayMedia.push({ id, type: \"map\", from: meta.title }));\n tables.slice(0, 2).forEach(id => autoDisplayMedia.push({ id, type: \"table\", from: meta.title }));\n });\n\n debugInfo += `Would auto-display ${autoDisplayMedia.length} media items:\\n`;\n autoDisplayMedia.slice(0, 10).forEach(m => {\n debugInfo += ` - ${m.type} ${m.id} from \"${m.from.substring(0, 50)}...\"\\n`;\n });\n } catch (searchError) {\n debugInfo += `❌ Search Error: ${searchError.message}\\n`;\n }\n\n return debugInfo;\n}\n\n// Debug endpoint for testing dates\nasync function handleDebugDates() {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) => new Date(b.date_iso || b.date).getTime() - new Date(a.date_iso || a.date).getTime());\n\n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n\n // Format the response\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10:\\n`;\n\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return summary + formattedArticles;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoints\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n const debugInfo = await handleDebugQuery(testQuery);\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n if (message === \"DEBUG_DATES\") {\n const debugInfo = await handleDebugDates();\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n // Verify required environment variables\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // OpenAI API headers\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n };\n\n console.log(\"Processing message:\", message.substring(0, 100) + \"...\");\n\n // Parse the query first (if parser assistant is configured)\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n console.log(\"Parsing query with fast assistant...\");\n parsedIntent = await parseQueryIntent(message);\n console.log(\"Parsed intent:\", parsedIntent);\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n }\n\n console.log(`Search strategy: ${searchStrategy}`);\n\n // Build filters from parsed intent\n let filters = {};\n if (parsedIntent?.explicit_filters?.company_only) {\n filters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n // Create or retrieve thread\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n\n // Add context based on parsed intent\n if (parsedIntent?.temporal?.tense === \"present\") {\n enhancedMessage +=\n \"\\n\\n[System Note: User query is present-tense. Prioritize recent data (marked as PRIMARY_SOURCE) for current facts and figures. Use older articles (marked as HISTORICAL_CONTEXT) only for background and trends.]\";\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from search results are automatically displayed to the user in a side panel.]\";\n }\n\n // Add message to thread\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[], context: any) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n\n // Use the context passed to this function\n const searchQuery = parsedArgs.query;\n const searchFilters = { ...context.filters, ...(parsedArgs.filters || {}) };\n\n // Get results\n const results = await searchPinecone(searchQuery, searchFilters, 50);\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Truncate text for OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500;\n if (textContent.length > maxTextLength) {\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date - newest first\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // Apply search strategy\n let finalResults = formattedResults;\n\n if (context.searchStrategy === \"recent_focused\") {\n // For present-tense queries, prioritize recent data\n const sixMonthsAgo = new Date();\n sixMonthsAgo.setMonth(sixMonthsAgo.getMonth() - 6);\n\n const recentResults = formattedResults.filter(r => new Date(r.publish_date_iso || r.date) > sixMonthsAgo);\n const historicalResults = formattedResults.filter(r =>\n new Date(r.publish_date_iso || r.date) <= sixMonthsAgo\n ).slice(0, 5);\n\n // Mark results with their context\n recentResults.forEach(r => r._context = \"PRIMARY_SOURCE\");\n historicalResults.forEach(r => r._context = \"HISTORICAL_CONTEXT\");\n\n // Combine with recent first\n finalResults = [...recentResults.slice(0, 20), ...historicalResults];\n }\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n\n // Use top results for media display\n let resultsToScan = finalResults\n .filter(r => !r._context || r._context === \"PRIMARY_SOURCE\")\n .slice(0, 10);\n\n if (resultsToScan.length < 5) {\n resultsToScan = finalResults.slice(0, 10);\n }\n\n resultsToScan.forEach((result) => {\n // Add all media types from results\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(`Returning ${finalResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(finalResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders },\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n // Pass context to handleToolCalls\n const context = {\n filters,\n searchStrategy,\n parsedIntent,\n };\n\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls, context);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders },\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",28,[142],{"_143":144,"_99":221,"_104":222,"_146":132,"_161":223},39,["D",1748686958767],[224],{"_102":225,"_146":132,"_143":144,"_95":142,"_154":155,"_152":153,"_99":226,"_145":-5,"_104":222,"_147":221,"_157":-5,"_148":27,"_156":27,"_98":227},"2d851d24-3e09-11f0-9121-76b3cceeab13",27,[142],{"_143":144,"_99":229,"_104":230,"_146":132,"_161":231},38,["D",1748686695388],[232],{"_102":233,"_146":132,"_143":144,"_95":142,"_154":234,"_152":153,"_99":235,"_145":-5,"_104":230,"_147":229,"_157":-5,"_148":27,"_156":27,"_98":236},"9088280e-3e08-11f0-a191-76b3cceeab13","// Debug endpoint for testing specific queries - shows complete flow\nif (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n\n let debugInfo = `🔍 COMPLETE QUERY FLOW DEBUG for: \"${testQuery}\"\\n${\"=\".repeat(80)}\\n\\n`;\n\n // Step 1: Parse the query\n debugInfo += `📝 STEP 1: QUERY PARSING\\n${\"-\".repeat(40)}\\n`;\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n try {\n parsedIntent = await parseQueryIntent(testQuery);\n debugInfo += `✅ Parser Results:\\n${JSON.stringify(parsedIntent, null, 2)}\\n\\n`;\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n debugInfo += `📊 Search Strategy: ${searchStrategy}\\n\\n`;\n } catch (e) {\n debugInfo += `❌ Parser Error: ${e.message}\\n\\n`;\n }\n } else {\n debugInfo += `⚠️ Parser not configured (PARSER_ASSISTANT_ID missing)\\n\\n`;\n }\n\n // Step 2: Build filters\n debugInfo += `🔧 STEP 2: FILTER BUILDING\\n${\"-\".repeat(40)}\\n`;\n const baseFilters = {};\n if (mediaFilters?.requireCharts) baseFilters.has_charts = true;\n if (mediaFilters?.requireMaps) baseFilters.has_maps = true;\n if (mediaFilters?.requireTables) baseFilters.has_tables = true;\n\n // Apply explicit company filter if specified\n if (parsedIntent?.explicit_filters?.company_only) {\n baseFilters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n debugInfo += `📋 Final Filters for Pinecone:\\n${JSON.stringify(baseFilters, null, 2)}\\n\\n`;\n\n // Step 3: Run Pinecone search\n debugInfo += `🔎 STEP 3: PINECONE SEARCH\\n${\"-\".repeat(40)}\\n`;\n try {\n const searchResults = await searchPinecone(testQuery, baseFilters, 20);\n\n debugInfo += `✅ Found ${searchResults.length} results\\n\\n`;\n debugInfo += `📊 Top 5 Results Summary:\\n`;\n\n searchResults.slice(0, 5).forEach((result, i) => {\n const meta = result.metadata;\n debugInfo += `\\n[${i + 1}] Score: ${result.score.toFixed(4)}\\n`;\n debugInfo += ` Title: ${meta.title ? meta.title.substring(0, 80) : \"No title\"}\\n`;\n debugInfo += ` Date: ${meta.publish_date} | Article ID: ${meta.article_id}\\n`;\n debugInfo += ` Charts: ${meta.all_charts || \"[]\"}\\n`;\n debugInfo += ` Maps: ${meta.all_maps || \"[]\"}\\n`;\n debugInfo += ` Tables: ${meta.all_tables || \"[]\"}\\n`;\n debugInfo += ` Companies: ${meta.companies || \"[]\"}\\n`;\n });\n\n // Step 4: Format data for OpenAI\n debugInfo += `\\n\\n📦 STEP 4: DATA FORMATTING FOR OPENAI\\n${\"-\".repeat(40)}\\n`;\n\n const formattedResults = searchResults.slice(0, 10).map((match) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Truncate text for OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500;\n if (textContent.length > maxTextLength) {\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n };\n });\n\n // Sort by date\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n debugInfo += `✅ Formatted ${formattedResults.length} results for OpenAI\\n`;\n debugInfo += `📄 Sample formatted result:\\n${\n JSON.stringify(formattedResults[0], null, 2).substring(0, 800)\n }...\\n\\n`;\n\n // Step 5: Prepare message for assistant\n debugInfo += `💬 STEP 5: MESSAGE PREPARATION FOR ASSISTANT\\n${\"-\".repeat(40)}\\n`;\n\n let enhancedMessage = testQuery;\n if (parsedIntent?.temporal?.tense === \"present\") {\n enhancedMessage +=\n \"\\n\\n[System Note: User query is present-tense. Prioritize recent data (marked as PRIMARY_SOURCE) for current facts and figures. Use older articles (marked as HISTORICAL_CONTEXT) only for background and trends.]\";\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from search results are automatically displayed to the user in a side panel.]\";\n }\n\n debugInfo += `📝 Enhanced Message:\\n\"${enhancedMessage}\"\\n\\n`;\n\n // Step 6: Simulate what would be sent to assistant\n debugInfo += `🤖 STEP 6: ASSISTANT INTERACTION SUMMARY\\n${\"-\".repeat(40)}\\n`;\n debugInfo += `Assistant ID: ${ASSISTANT_ID}\\n`;\n debugInfo += `Would create/use thread and send:\\n`;\n debugInfo += `1. User message: \"${testQuery}\"\\n`;\n debugInfo += `2. Tool calls would include:\\n`;\n debugInfo += ` - search_articles(query=\"${testQuery}\", filters=${JSON.stringify(baseFilters)})\\n`;\n debugInfo += `3. Tool would return ${formattedResults.length} formatted results\\n`;\n debugInfo += `4. Assistant would process and generate response\\n\\n`;\n\n // Media auto-display summary\n debugInfo += `🖼️ MEDIA AUTO-DISPLAY\\n${\"-\".repeat(40)}\\n`;\n const autoDisplayMedia = [];\n formattedResults.slice(0, 5).forEach((result) => {\n result.all_charts.forEach(id => autoDisplayMedia.push({ id, type: \"chart\", from: result.title }));\n result.all_maps.forEach(id => autoDisplayMedia.push({ id, type: \"map\", from: result.title }));\n result.all_tables.slice(0, 2).forEach(id => autoDisplayMedia.push({ id, type: \"table\", from: result.title }));\n });\n\n debugInfo += `Would auto-display ${autoDisplayMedia.length} media items:\\n`;\n autoDisplayMedia.slice(0, 10).forEach(m => {\n debugInfo += ` - ${m.type} ${m.id} from \"${m.from.substring(0, 50)}...\"\\n`;\n });\n } catch (searchError) {\n debugInfo += `❌ Search Error: ${searchError.message}\\n`;\n }\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n}",26,[142],{"_143":144,"_99":238,"_104":239,"_146":132,"_161":240},37,["D",1748686647458],[241],{"_102":242,"_146":132,"_143":144,"_95":142,"_154":234,"_152":153,"_99":243,"_145":-5,"_104":239,"_147":238,"_157":-5,"_148":27,"_156":27,"_98":244},"73f6dd98-3e08-11f0-a1e1-76b3cceeab13",25,[142],{"_143":144,"_99":246,"_104":247,"_146":132,"_161":248},36,["D",1748678938332],[249],{"_102":250,"_146":132,"_143":144,"_95":142,"_154":251,"_152":153,"_99":252,"_145":-5,"_104":247,"_147":246,"_157":-5,"_148":27,"_156":27,"_98":253},"80f805d8-3df6-11f0-a1db-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n// - PARSER_ASSISTANT_ID (optional)\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\nconst PARSER_ASSISTANT_ID = Deno.env.get(\"PARSER_ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Fast query parser using GPT-3.5\nasync function parseQueryIntent(query: string) {\n try {\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n\n // Create a quick thread for parsing\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({}),\n });\n\n const thread = await threadResponse.json();\n\n // Add the query\n await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n role: \"user\",\n content: query,\n }),\n });\n\n // Run the parser assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({\n assistant_id: PARSER_ASSISTANT_ID,\n max_completion_tokens: 200,\n }),\n });\n\n const run = await runResponse.json();\n\n // Poll for completion\n let runStatus = run;\n const maxWaitTime = 5000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Parser timeout\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 500));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n runStatus = await statusResponse.json();\n }\n\n // Get the response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=1`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n },\n );\n\n const messages = await messagesResponse.json();\n const assistantMessage = messages.data[0];\n\n if (assistantMessage && assistantMessage.content && assistantMessage.content[0]) {\n const content = assistantMessage.content[0].text.value;\n return JSON.parse(content);\n }\n\n throw new Error(\"No parser response\");\n } catch (error) {\n console.error(\"Parser error:\", error);\n // Return default if parser fails\n return {\n temporal: {\n tense: \"present\",\n explicit_date: null,\n recency_signal: \"none\",\n },\n explicit_filters: {\n company_only: null,\n exclude: null,\n },\n };\n }\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 50) {\n const index = pinecone.index(\"mees-articles\");\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(\"Filters:\", JSON.stringify(filters));\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build Pinecone filter from provided filters\n let pineconeFilter = {};\n\n // Only apply explicit filters\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone - let semantic search do the heavy lifting\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0);\n\n try {\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId, mediaFilters = {} } = await req.json();\n\n // Debug endpoint for testing specific queries\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n\n let debugInfo = `Query Analysis for: \"${testQuery}\"\\n\\n`;\n\n // Test the parser if available\n if (PARSER_ASSISTANT_ID) {\n const parsedIntent = await parseQueryIntent(testQuery);\n debugInfo += `Parser Results:\\n${JSON.stringify(parsedIntent, null, 2)}\\n\\n`;\n } else {\n debugInfo += `Parser not configured (PARSER_ASSISTANT_ID missing)\\n\\n`;\n }\n\n // Show what filters would be applied\n const baseFilters = {};\n if (mediaFilters?.requireCharts) baseFilters.has_charts = true;\n if (mediaFilters?.requireMaps) baseFilters.has_maps = true;\n if (mediaFilters?.requireTables) baseFilters.has_tables = true;\n\n debugInfo += `Media Filters from UI: ${JSON.stringify(baseFilters, null, 2)}\\n`;\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n has_charts: meta.has_charts,\n has_maps: meta.has_maps,\n has_tables: meta.has_tables,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) =>\n new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime()\n );\n\n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n\n // Format the response with all metadata\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10 with full metadata:\\n`;\n\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n articleInfo += ` 🏷️ Has Media: Charts=${d.has_charts} Maps=${d.has_maps} Tables=${d.has_tables}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === \"string\") {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || \"N/A\"}: ${parsed.context || \"N/A\"}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === \"object\") {\n // If it's already an object\n articleInfo += ` • ${km.metric || \"N/A\"}: ${km.context || \"N/A\"}\\n`;\n }\n });\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return new Response(\n JSON.stringify({\n response: summary + formattedArticles,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Parse the query first (if parser assistant is configured)\n let parsedIntent = null;\n let searchStrategy = \"balanced\";\n\n if (PARSER_ASSISTANT_ID) {\n console.log(\"Parsing query with fast assistant...\");\n parsedIntent = await parseQueryIntent(message);\n console.log(\"Parsed intent:\", parsedIntent);\n\n // Determine search strategy\n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n }\n\n console.log(`Search strategy: ${searchStrategy}`);\n\n // Build base filters from UI toggles\n let baseFilters = {};\n if (mediaFilters.requireCharts) baseFilters.has_charts = true;\n if (mediaFilters.requireMaps) baseFilters.has_maps = true;\n if (mediaFilters.requireTables) baseFilters.has_tables = true;\n\n // Apply explicit company filter if specified\n if (parsedIntent?.explicit_filters?.company_only) {\n baseFilters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n\n // Add context based on parsed intent\n if (parsedIntent?.temporal?.tense === \"present\") {\n enhancedMessage +=\n \"\\n\\n[System Note: User query is present-tense. Prioritize recent data (marked as PRIMARY_SOURCE) for current facts and figures. Use older articles (marked as HISTORICAL_CONTEXT) only for background and trends.]\";\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from search results are automatically displayed to the user in a side panel.]\";\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[], context: any) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n\n // Use the context passed to this function\n const searchQuery = parsedArgs.query;\n const searchFilters = { ...context.baseFilters, ...(parsedArgs.filters || {}) };\n\n // Get more results initially\n const results = await searchPinecone(\n searchQuery,\n searchFilters,\n 50, // Get more results for filtering\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format and categorize results based on search strategy\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500;\n if (textContent.length > maxTextLength) {\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // Categorize results based on search strategy\n let finalResults = formattedResults;\n\n if (context.searchStrategy === \"recent_focused\") {\n // For present-tense queries, prioritize recent data\n const sixMonthsAgo = new Date();\n sixMonthsAgo.setMonth(sixMonthsAgo.getMonth() - 6);\n\n const recentResults = formattedResults.filter(r => new Date(r.publish_date_iso || r.date) > sixMonthsAgo);\n\n const historicalResults = formattedResults.filter(r =>\n new Date(r.publish_date_iso || r.date) <= sixMonthsAgo\n ).slice(0, 5);\n\n // Mark results with their context\n recentResults.forEach(r => r._context = \"PRIMARY_SOURCE\");\n historicalResults.forEach(r => r._context = \"HISTORICAL_CONTEXT\");\n\n // Combine with recent first\n finalResults = [...recentResults.slice(0, 20), ...historicalResults];\n }\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n\n // Use top results for media display (prioritize PRIMARY_SOURCE)\n let resultsToScan = finalResults\n .filter(r => !r._context || r._context === \"PRIMARY_SOURCE\")\n .slice(0, 10);\n\n if (resultsToScan.length < 5) {\n resultsToScan = finalResults.slice(0, 10);\n }\n\n resultsToScan.forEach((result, index) => {\n // Add all media types from results\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(\n \"Results after sorting:\",\n finalResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n context: r._context || \"STANDARD\",\n })),\n );\n\n console.log(`Returning ${finalResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(finalResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000;\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders },\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n // Pass context to handleToolCalls\n const context = {\n baseFilters,\n searchStrategy,\n parsedIntent,\n };\n\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls, context);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders },\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",24,[142],{"_143":144,"_99":255,"_104":256,"_146":132,"_161":257},35,["D",1748677978907],[258],{"_102":259,"_146":132,"_143":144,"_95":142,"_154":260,"_152":153,"_99":261,"_145":-5,"_104":256,"_147":255,"_157":-5,"_148":27,"_156":27,"_98":262},"451b1138-3df4-11f0-85ae-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\")\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\")\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\nconst PARSER_ASSISTANT_ID = Deno.env.get(\"PARSER_ASSISTANT_ID\"); // Add this to Val settings\n\n// Fast query parser using GPT-3.5\nasync function parseQueryIntent(query: string) {\n try {\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n \n // Create a quick thread for parsing\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\"\n },\n body: JSON.stringify({})\n });\n \n const thread = await threadResponse.json();\n \n // Add the query\n await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\"\n },\n body: JSON.stringify({\n role: \"user\",\n content: query\n })\n });\n \n // Run the parser assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\"\n },\n body: JSON.stringify({\n assistant_id: PARSER_ASSISTANT_ID,\n max_completion_tokens: 200 // Keep it short\n })\n });\n \n const run = await runResponse.json();\n \n // Poll for completion (should be fast with GPT-3.5)\n let runStatus = run;\n const maxWaitTime = 5000; // 5 seconds max for parser\n const startTime = Date.now();\n \n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Parser timeout\");\n }\n \n await new Promise(resolve => setTimeout(resolve, 500));\n \n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\"\n }\n }\n );\n \n runStatus = await statusResponse.json();\n }\n \n // Get the response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=1`,\n {\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"OpenAI-Beta\": \"assistants=v2\"\n }\n }\n );\n \n const messages = await messagesResponse.json();\n const assistantMessage = messages.data[0];\n \n if (assistantMessage && assistantMessage.content && assistantMessage.content[0]) {\n const content = assistantMessage.content[0].text.value;\n return JSON.parse(content);\n }\n \n throw new Error(\"No parser response\");\n \n } catch (error) {\n console.error(\"Parser error:\", error);\n // Return default if parser fails\n return {\n temporal: {\n tense: \"present\",\n explicit_date: null,\n recency_signal: \"none\"\n },\n explicit_filters: {\n company_only: null,\n exclude: null\n }\n };\n }\n}\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 50) {\n const index = pinecone.index(\"mees-articles\");\n \n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(\"Filters:\", JSON.stringify(filters));\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build Pinecone filter from provided filters\n let pineconeFilter = {};\n \n // Only apply explicit filters\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n \n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n \n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n \n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n \n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone - let semantic search do the heavy lifting\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n \n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n \n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(' ');\n const [month, day, year] = datePart.split('/');\n const [hour, minute] = timePart ? timePart.split(':') : ['0', '0'];\n \n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n \n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId, mediaFilters = {} } = await req.json();\n \n // Debug endpoint for testing specific queries\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n \n let debugInfo = `Query Analysis for: \"${testQuery}\"\\n\\n`;\n \n // Test the parser if available\n if (PARSER_ASSISTANT_ID) {\n const parsedIntent = await parseQueryIntent(testQuery);\n debugInfo += `Parser Results:\\n${JSON.stringify(parsedIntent, null, 2)}\\n\\n`;\n } else {\n debugInfo += `Parser not configured (PARSER_ASSISTANT_ID missing)\\n\\n`;\n }\n \n // Show what filters would be applied\n const baseFilters = {};\n if (mediaFilters?.requireCharts) baseFilters.has_charts = true;\n if (mediaFilters?.requireMaps) baseFilters.has_maps = true;\n if (mediaFilters?.requireTables) baseFilters.has_tables = true;\n \n debugInfo += `Media Filters from UI: ${JSON.stringify(baseFilters, null, 2)}\\n`;\n \n return new Response(JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\"\n }), { \n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\"\n }\n });\n }\n \n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n \n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n \n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n has_charts: meta.has_charts,\n has_maps: meta.has_maps,\n has_tables: meta.has_tables\n };\n });\n \n // Sort by ISO date\n enrichedData.sort((a, b) => new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime());\n \n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n \n // Format the response with all metadata\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10 with full metadata:\\n`;\n \n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i+1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || 'missing'}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || 'N/A'} | Section: ${d.section || 'N/A'}\\n`;\n articleInfo += ` 🏷️ Has Media: Charts=${d.has_charts} Maps=${d.has_maps} Tables=${d.has_tables}\\n`;\n \n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(', ')}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(', ')}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(', ')}\\n`;\n }\n \n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(', ')}\\n`;\n }\n \n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === 'string') {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || 'N/A'}: ${parsed.context || 'N/A'}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === 'object') {\n // If it's already an object\n articleInfo += ` • ${km.metric || 'N/A'}: ${km.context || 'N/A'}\\n`;\n }\n });\n }\n \n return articleInfo;\n }).join('\\n' + '-'.repeat(60) + '\\n');\n \n return new Response(JSON.stringify({\n response: summary + formattedArticles,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\"\n }), { \n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\"\n }\n });\n }\n \n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n \n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n \n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\" // This is the critical header for v2 API\n };\n \n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n \n // Parse the query first (if parser assistant is configured)\n let parsedIntent = null;\n if (PARSER_ASSISTANT_ID) {\n console.log(\"Parsing query with fast assistant...\");\n parsedIntent = await parseQueryIntent(message);\n console.log(\"Parsed intent:\", parsedIntent);\n }\n \n // Enhanced search based on parsed intent\n let searchResults = {\n current_data: [],\n historical_context: []\n };\n \n // Build base filters from UI toggles\n let baseFilters = {};\n if (mediaFilters.requireCharts) baseFilters.has_charts = true;\n if (mediaFilters.requireMaps) baseFilters.has_maps = true;\n if (mediaFilters.requireTables) baseFilters.has_tables = true;\n \n // Apply explicit company filter if specified\n if (parsedIntent?.explicit_filters?.company_only) {\n baseFilters.companies = { $in: [parsedIntent.explicit_filters.company_only] };\n }\n \n // Determine search strategy based on parsed intent\n const now = new Date();\n let searchStrategy = \"balanced\"; // default\n \n if (parsedIntent?.temporal?.tense === \"present\" && !parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"recent_focused\";\n } else if (parsedIntent?.temporal?.explicit_date) {\n searchStrategy = \"date_specific\";\n }\n \n console.log(`Search strategy: ${searchStrategy}`);\n \n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({})\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n \n // Add context based on parsed intent\n if (parsedIntent?.temporal?.tense === \"present\") {\n enhancedMessage += \"\\n\\n[System Note: User query is present-tense. Prioritize recent data (marked as PRIMARY_SOURCE) for current facts and figures. Use older articles (marked as HISTORICAL_CONTEXT) only for background and trends.]\";\n } else {\n enhancedMessage += \"\\n\\n[System Note: Charts, maps, and tables from search results are automatically displayed to the user in a side panel.]\";\n }\n \n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage\n })\n });\n \n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID\n })\n });\n \n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n \n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n \n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n \n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n \n // Use the search strategy from earlier\n const searchQuery = parsedArgs.query;\n const searchFilters = { ...baseFilters, ...(parsedArgs.filters || {}) };\n \n // Get more results initially\n const results = await searchPinecone(\n searchQuery,\n searchFilters,\n 50 // Get more results for filtering\n );\n \n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format and categorize results based on search strategy\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n \n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500; // Limit text sent to OpenAI\n if (textContent.length > maxTextLength) {\n // Keep the beginning which usually has the most important info\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent, // Now truncated for faster processing\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country\n };\n });\n \n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n \n // Categorize results based on search strategy\n let finalResults = formattedResults;\n \n if (searchStrategy === \"recent_focused\") {\n // For present-tense queries, prioritize recent data\n const sixMonthsAgo = new Date();\n sixMonthsAgo.setMonth(sixMonthsAgo.getMonth() - 6);\n \n const recentResults = formattedResults.filter(r => \n new Date(r.publish_date_iso || r.date) > sixMonthsAgo\n );\n \n const historicalResults = formattedResults.filter(r => \n new Date(r.publish_date_iso || r.date) <= sixMonthsAgo\n ).slice(0, 5); // Just top 5 for context\n \n // Mark results with their context\n recentResults.forEach(r => r._context = \"PRIMARY_SOURCE\");\n historicalResults.forEach(r => r._context = \"HISTORICAL_CONTEXT\");\n \n // Combine with recent first\n finalResults = [...recentResults.slice(0, 20), ...historicalResults];\n }\n \n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n \n // Use top results for media display (prioritize PRIMARY_SOURCE)\n let resultsToScan = finalResults\n .filter(r => !r._context || r._context === \"PRIMARY_SOURCE\")\n .slice(0, 10);\n \n if (resultsToScan.length < 5) {\n // If not enough recent results, include some historical\n resultsToScan = finalResults.slice(0, 10);\n }\n \n resultsToScan.forEach((result, index) => {\n // Add all media types from results\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n \n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n \n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n });\n \n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n \n console.log(\"Results after sorting:\", finalResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n context: r._context || \"STANDARD\"\n })));\n\n console.log(`Returning ${finalResults.length} formatted results`);\n \n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(finalResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n \n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n \n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders } // v2 headers\n );\n \n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n \n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n \n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n \n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs })\n }\n );\n \n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders } // v2 headers\n );\n \n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n \n const messages = await messagesResponse.json();\n \n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n \n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n \n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n \n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n \n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",23,[142],{"_143":144,"_99":264,"_104":265,"_146":132,"_161":266},34,["D",1748676540976],[267],{"_102":268,"_146":132,"_143":144,"_95":142,"_154":269,"_152":153,"_99":270,"_145":-5,"_104":265,"_147":264,"_157":-5,"_148":27,"_156":27,"_98":271},"ec07e7fe-3df0-11f0-84b8-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\")\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\")\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string, queryIntent: any = null) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n \n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true\n };\n }\n \n // Check for specific months\n const monthYearMatch = query.match(/\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i);\n if (monthYearMatch) {\n const monthNames = [\"january\", \"february\", \"march\", \"april\", \"may\", \"june\", \"july\", \"august\", \"september\", \"october\", \"november\", \"december\"];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true\n };\n }\n \n // Check for relative time indicators\n if (lowerQuery.includes('latest') || lowerQuery.includes('most recent') || lowerQuery.includes('newest') || lowerQuery.includes('current')) {\n // If searching for specific media (maps, charts, tables), use longer timeframe\n // since media items are less frequent than articles\n let monthsBack = 1; // Default for general queries\n \n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6; // Look back 6 months for media items\n }\n \n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n searchingForMedia: queryIntent && queryIntent.wantsMedia\n };\n }\n \n if (lowerQuery.includes('recent')) {\n // Last 3 months for \"recent\" (or 6 for media)\n let monthsBack = 3;\n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6;\n }\n \n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false\n };\n }\n \n if (lowerQuery.includes('this year')) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true\n };\n }\n \n if (lowerQuery.includes('last year')) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true\n };\n }\n \n // Default: Last 3 months for general queries, 6 months for media\n let monthsBack = 3;\n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6;\n }\n \n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true\n };\n}\n\n// Analyze query for specific search intents (media, companies, etc.)\nfunction analyzeQueryIntent(query: string) {\n const lowerQuery = query.toLowerCase();\n const intent = {\n wantsCharts: false,\n wantsMaps: false,\n wantsTables: false,\n wantsMedia: false,\n companies: [],\n countries: []\n };\n \n // Check for media requests\n if (lowerQuery.includes('chart') || lowerQuery.includes('graph') || lowerQuery.includes('visualiz')) {\n intent.wantsCharts = true;\n intent.wantsMedia = true;\n }\n \n if (lowerQuery.includes('map') || lowerQuery.includes('geography') || lowerQuery.includes('location')) {\n intent.wantsMaps = true;\n intent.wantsMedia = true;\n }\n \n if (lowerQuery.includes('table') || lowerQuery.includes('data table')) {\n intent.wantsTables = true;\n intent.wantsMedia = true;\n }\n \n // Check for country mentions\n const countries = ['egypt', 'saudi', 'uae', 'qatar', 'kuwait', 'iraq', 'iran', 'israel', 'jordan', 'lebanon', 'syria', 'turkey', 'cyprus'];\n countries.forEach(country => {\n if (lowerQuery.includes(country)) {\n // Capitalize first letter for proper matching\n let properCountry = country.charAt(0).toUpperCase() + country.slice(1);\n // Fix common misspellings in metadata\n if (properCountry === 'Egypt') {\n intent.countries.push('Egypt', 'Eygpt'); // Include misspelling found in data\n } else {\n intent.countries.push(properCountry);\n }\n }\n });\n \n // Check for company mentions (add more as needed)\n const companies = ['chevron', 'exxon', 'bp', 'shell', 'eni', 'total', 'apache', 'wintershall'];\n companies.forEach(company => {\n if (lowerQuery.includes(company)) {\n intent.companies.push(company);\n }\n });\n \n return intent;\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n \n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n \n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n \n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n \n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n \n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n \n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n \n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined\n });\n \n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n \n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n \n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n \n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n \n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n \n // Analyze query intent for media/companies/countries FIRST\n const queryIntent = analyzeQueryIntent(query);\n console.log(\"Query intent analysis:\", queryIntent);\n \n // Analyze query for temporal indicators, passing the intent\n const dateRange = analyzeQueryTemporal(query, queryIntent);\n console.log(\"Date range analysis:\", dateRange);\n \n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n \n // Build filter object for Pinecone\n let pineconeFilter = {};\n \n // Add intent-based filters FIRST (these take priority)\n if (queryIntent.wantsCharts) {\n pineconeFilter.has_charts = true;\n }\n \n if (queryIntent.wantsMaps) {\n pineconeFilter.has_maps = true;\n }\n \n if (queryIntent.wantsTables) {\n pineconeFilter.has_tables = true;\n }\n \n // Add country filter from intent or explicit filter\n if (queryIntent.countries.length > 0) {\n // For Egypt, handle the misspelling in the data\n if (queryIntent.countries.includes('Egypt')) {\n // Can't use $in with primary_country, so just use the misspelling\n pineconeFilter.primary_country = 'Eygpt'; // Use the misspelling that's in the data\n } else {\n pineconeFilter.primary_country = queryIntent.countries[0];\n }\n } else if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n \n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n } else if (queryIntent.companies.length > 0) {\n pineconeFilter.companies = { $in: queryIntent.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n \n // If we have specific media requests, increase topK to ensure we get enough results\n if (queryIntent.wantsMedia) {\n topK = Math.max(topK, 30);\n }\n \n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n \n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n \n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n \n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n \n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n \n console.log(`${searchResults.matches.length} results after date filtering`);\n \n // If searching for specific media, boost scores for articles with more of that media type\n if (queryIntent.wantsMedia) {\n searchResults.matches.forEach(match => {\n let mediaBoost = 1.0;\n \n if (queryIntent.wantsCharts) {\n const charts = JSON.parse(match.metadata.all_charts || \"[]\");\n if (charts.length > 0) mediaBoost *= (1 + charts.length * 0.1);\n }\n \n if (queryIntent.wantsMaps) {\n const maps = JSON.parse(match.metadata.all_maps || \"[]\");\n if (maps.length > 0) mediaBoost *= (1 + maps.length * 0.2); // Maps are rarer, so bigger boost\n }\n \n if (queryIntent.wantsTables) {\n const tables = JSON.parse(match.metadata.all_tables || \"[]\");\n if (tables.length > 0) mediaBoost *= (1 + tables.length * 0.1);\n }\n \n match.score *= mediaBoost;\n });\n \n // Re-sort by boosted scores\n searchResults.matches.sort((a, b) => b.score - a.score);\n }\n \n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n \n // If we have metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n \n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n \n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n \n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(' ');\n const [month, day, year] = datePart.split('/');\n const [hour, minute] = timePart ? timePart.split(':') : ['0', '0'];\n \n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n \n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n \n // Debug endpoint for testing specific queries\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n const queryIntent = analyzeQueryIntent(testQuery);\n const dateRange = analyzeQueryTemporal(testQuery, queryIntent);\n const metrics = extractMetricsFromQuery(testQuery);\n \n let debugInfo = `Query Analysis for: \"${testQuery}\"\\n\\n`;\n debugInfo += `Intent Analysis:\\n`;\n debugInfo += ` - Wants Charts: ${queryIntent.wantsCharts}\\n`;\n debugInfo += ` - Wants Maps: ${queryIntent.wantsMaps}\\n`;\n debugInfo += ` - Wants Tables: ${queryIntent.wantsTables}\\n`;\n debugInfo += ` - Wants Media: ${queryIntent.wantsMedia}\\n`;\n debugInfo += ` - Countries: ${queryIntent.countries.join(', ') || 'None'}\\n`;\n debugInfo += ` - Companies: ${queryIntent.companies.join(', ') || 'None'}\\n\\n`;\n \n debugInfo += `Date Range Analysis:\\n`;\n debugInfo += ` - Start Date: ${dateRange.startDate}\\n`;\n debugInfo += ` - End Date: ${dateRange.endDate}\\n`;\n debugInfo += ` - Is Latest: ${dateRange.isLatest || false}\\n`;\n debugInfo += ` - Is Default: ${dateRange.isDefault || false}\\n`;\n debugInfo += ` - Searching for Media: ${dateRange.searchingForMedia || false}\\n\\n`;\n \n debugInfo += `Extracted Metrics: ${metrics.join(', ') || 'None'}\\n\\n`;\n \n // Build the filter that would be used\n let pineconeFilter = {};\n if (queryIntent.wantsCharts) pineconeFilter.has_charts = true;\n if (queryIntent.wantsMaps) pineconeFilter.has_maps = true;\n if (queryIntent.wantsTables) pineconeFilter.has_tables = true;\n if (queryIntent.countries.includes('Egypt')) {\n pineconeFilter.primary_country = 'Eygpt'; // Handle misspelling\n }\n \n debugInfo += `Pinecone Filter: ${JSON.stringify(pineconeFilter, null, 2)}\\n`;\n \n return new Response(JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\"\n }), { \n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\"\n }\n });\n }\n \n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n \n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n \n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n has_charts: meta.has_charts,\n has_maps: meta.has_maps,\n has_tables: meta.has_tables\n };\n });\n \n // Sort by ISO date\n enrichedData.sort((a, b) => new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime());\n \n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n \n // Format the response with all metadata\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10 with full metadata:\\n`;\n \n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i+1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || 'missing'}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || 'N/A'} | Section: ${d.section || 'N/A'}\\n`;\n articleInfo += ` 🏷️ Has Media: Charts=${d.has_charts} Maps=${d.has_maps} Tables=${d.has_tables}\\n`;\n \n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(', ')}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(', ')}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(', ')}\\n`;\n }\n \n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(', ')}\\n`;\n }\n \n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === 'string') {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || 'N/A'}: ${parsed.context || 'N/A'}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === 'object') {\n // If it's already an object\n articleInfo += ` • ${km.metric || 'N/A'}: ${km.context || 'N/A'}\\n`;\n }\n });\n }\n \n return articleInfo;\n }).join('\\n' + '-'.repeat(60) + '\\n');\n \n return new Response(JSON.stringify({\n response: summary + formattedArticles,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\"\n }), { \n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\"\n }\n });\n }\n \n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n \n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n \n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\" // This is the critical header for v2 API\n };\n \n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n \n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({})\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n const queryIntent = analyzeQueryIntent(message);\n \n if (queryIntent.wantsMedia) {\n let mediaContext = \"\\n\\n[System Note: User is looking for \";\n const mediaTypes = [];\n if (queryIntent.wantsCharts) mediaTypes.push(\"charts\");\n if (queryIntent.wantsMaps) mediaTypes.push(\"maps\");\n if (queryIntent.wantsTables) mediaTypes.push(\"tables\");\n mediaContext += mediaTypes.join(\" and \");\n mediaContext += \". Relevant media from search results is automatically displayed. Focus on describing what these visualizations show.]\";\n enhancedMessage += mediaContext;\n } else {\n enhancedMessage += \"\\n\\n[System Note: Charts, maps, and tables from the top search results are automatically displayed to the user in a side panel. Focus on analyzing the data and insights rather than listing media IDs.]\";\n }\n \n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage\n })\n });\n \n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID\n })\n });\n \n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n \n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n \n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n \n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n \n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n \n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500; // Limit text sent to OpenAI\n if (textContent.length > maxTextLength) {\n // Keep the beginning which usually has the most important info\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent, // Now truncated for faster processing\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country\n };\n });\n \n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n \n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n \n // If user is looking for specific media types, prioritize those\n const queryIntent = analyzeQueryIntent(parsedArgs.query);\n let resultsToScan = formattedResults.slice(0, 5); // Default: top 5\n \n if (queryIntent.wantsMedia) {\n // If looking for specific media, scan more results and filter\n resultsToScan = formattedResults.slice(0, 10);\n \n // Filter to only articles that have the requested media type\n if (queryIntent.wantsCharts) {\n resultsToScan = resultsToScan.filter(r => r.all_charts.length > 0);\n }\n if (queryIntent.wantsMaps) {\n resultsToScan = resultsToScan.filter(r => r.all_maps.length > 0);\n }\n if (queryIntent.wantsTables) {\n resultsToScan = resultsToScan.filter(r => r.all_tables.length > 0);\n }\n }\n \n resultsToScan.forEach((result, index) => {\n // Add charts if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsCharts) {\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n }\n \n // Add maps if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsMaps) {\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n }\n \n // Add tables if requested or by default (limit to first 2 per article)\n if (!queryIntent.wantsMedia || queryIntent.wantsTables) {\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n }\n });\n \n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n \n console.log(\"Results after sorting:\", formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score\n })));\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n \n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n \n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n \n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders } // v2 headers\n );\n \n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n \n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n \n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n \n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs })\n }\n );\n \n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders } // v2 headers\n );\n \n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n \n const messages = await messagesResponse.json();\n \n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n \n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n \n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n \n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n \n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",22,[142],{"_143":144,"_99":273,"_104":274,"_146":132,"_161":275},33,["D",1748615501659],[276],{"_102":277,"_146":132,"_143":144,"_95":142,"_154":278,"_152":153,"_99":279,"_145":-5,"_104":274,"_147":273,"_157":-5,"_148":27,"_156":27,"_98":280},"cdc34038-3d62-11f0-a0b4-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string, queryIntent: any = null) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n\n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for specific months\n const monthYearMatch = query.match(\n /\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i,\n );\n if (monthYearMatch) {\n const monthNames = [\n \"january\",\n \"february\",\n \"march\",\n \"april\",\n \"may\",\n \"june\",\n \"july\",\n \"august\",\n \"september\",\n \"october\",\n \"november\",\n \"december\",\n ];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for relative time indicators\n if (\n lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n ) {\n // If searching for specific media (maps, charts, tables), use longer timeframe\n // since media items are less frequent than articles\n let monthsBack = 1; // Default for general queries\n\n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6; // Look back 6 months for media items\n }\n\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n searchingForMedia: queryIntent && queryIntent.wantsMedia,\n };\n }\n\n if (lowerQuery.includes(\"recent\")) {\n // Last 3 months for \"recent\" (or 6 for media)\n let monthsBack = 3;\n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6;\n }\n\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n };\n }\n\n if (lowerQuery.includes(\"this year\")) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n if (lowerQuery.includes(\"last year\")) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Default: Last 3 months for general queries, 6 months for media\n let monthsBack = 3;\n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6;\n }\n\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true,\n };\n}\n\n// Analyze query for specific search intents (media, companies, etc.)\nfunction analyzeQueryIntent(query: string) {\n const lowerQuery = query.toLowerCase();\n const intent = {\n wantsCharts: false,\n wantsMaps: false,\n wantsTables: false,\n wantsMedia: false,\n companies: [],\n countries: [],\n };\n\n // Check for media requests\n if (lowerQuery.includes(\"chart\") || lowerQuery.includes(\"graph\") || lowerQuery.includes(\"visualiz\")) {\n intent.wantsCharts = true;\n intent.wantsMedia = true;\n }\n\n if (lowerQuery.includes(\"map\") || lowerQuery.includes(\"geography\") || lowerQuery.includes(\"location\")) {\n intent.wantsMaps = true;\n intent.wantsMedia = true;\n }\n\n if (lowerQuery.includes(\"table\") || lowerQuery.includes(\"data table\")) {\n intent.wantsTables = true;\n intent.wantsMedia = true;\n }\n\n // Check for country mentions\n const countries = [\n \"egypt\",\n \"saudi\",\n \"uae\",\n \"qatar\",\n \"kuwait\",\n \"iraq\",\n \"iran\",\n \"israel\",\n \"jordan\",\n \"lebanon\",\n \"syria\",\n \"turkey\",\n \"cyprus\",\n ];\n countries.forEach(country => {\n if (lowerQuery.includes(country)) {\n // Capitalize first letter for proper matching\n let properCountry = country.charAt(0).toUpperCase() + country.slice(1);\n // Fix common misspellings in metadata\n if (properCountry === \"Egypt\") {\n intent.countries.push(\"Egypt\", \"Eygpt\"); // Include misspelling found in data\n } else {\n intent.countries.push(properCountry);\n }\n }\n });\n\n // Check for company mentions (add more as needed)\n const companies = [\"chevron\", \"exxon\", \"bp\", \"shell\", \"eni\", \"total\", \"apache\", \"wintershall\"];\n companies.forEach(company => {\n if (lowerQuery.includes(company)) {\n intent.companies.push(company);\n }\n });\n\n return intent;\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n\n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n\n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n\n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n\n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n\n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n\n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n\n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined,\n });\n\n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n\n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n\n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Analyze query intent for media/companies/countries FIRST\n const queryIntent = analyzeQueryIntent(query);\n console.log(\"Query intent analysis:\", queryIntent);\n\n // Analyze query for temporal indicators, passing the intent\n const dateRange = analyzeQueryTemporal(query, queryIntent);\n console.log(\"Date range analysis:\", dateRange);\n\n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n\n // Build filter object for Pinecone\n let pineconeFilter = {};\n\n // Add intent-based filters FIRST (these take priority)\n if (queryIntent.wantsCharts) {\n pineconeFilter.has_charts = true;\n }\n\n if (queryIntent.wantsMaps) {\n pineconeFilter.has_maps = true;\n }\n\n if (queryIntent.wantsTables) {\n pineconeFilter.has_tables = true;\n }\n\n // Add country filter from intent or explicit filter\n if (queryIntent.countries.length > 0) {\n // For Egypt, handle the misspelling in the data\n if (queryIntent.countries.includes(\"Egypt\")) {\n // Can't use $in with primary_country, so just use the misspelling\n pineconeFilter.primary_country = \"Eygpt\"; // Use the misspelling that's in the data\n } else {\n pineconeFilter.primary_country = queryIntent.countries[0];\n }\n } else if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n } else if (queryIntent.companies.length > 0) {\n pineconeFilter.companies = { $in: queryIntent.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n // If we have specific media requests, increase topK to ensure we get enough results\n if (queryIntent.wantsMedia) {\n topK = Math.max(topK, 30);\n }\n\n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n\n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n console.log(`${searchResults.matches.length} results after date filtering`);\n\n // If searching for specific media, boost scores for articles with more of that media type\n if (queryIntent.wantsMedia) {\n searchResults.matches.forEach(match => {\n let mediaBoost = 1.0;\n\n if (queryIntent.wantsCharts) {\n const charts = JSON.parse(match.metadata.all_charts || \"[]\");\n if (charts.length > 0) mediaBoost *= 1 + charts.length * 0.1;\n }\n\n if (queryIntent.wantsMaps) {\n const maps = JSON.parse(match.metadata.all_maps || \"[]\");\n if (maps.length > 0) mediaBoost *= 1 + maps.length * 0.2; // Maps are rarer, so bigger boost\n }\n\n if (queryIntent.wantsTables) {\n const tables = JSON.parse(match.metadata.all_tables || \"[]\");\n if (tables.length > 0) mediaBoost *= 1 + tables.length * 0.1;\n }\n\n match.score *= mediaBoost;\n });\n\n // Re-sort by boosted scores\n searchResults.matches.sort((a, b) => b.score - a.score);\n }\n\n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n\n // If we have metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n\n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing specific queries\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n const queryIntent = analyzeQueryIntent(testQuery);\n const dateRange = analyzeQueryTemporal(testQuery, queryIntent);\n const metrics = extractMetricsFromQuery(testQuery);\n\n let debugInfo = `Query Analysis for: \"${testQuery}\"\\n\\n`;\n debugInfo += `Intent Analysis:\\n`;\n debugInfo += ` - Wants Charts: ${queryIntent.wantsCharts}\\n`;\n debugInfo += ` - Wants Maps: ${queryIntent.wantsMaps}\\n`;\n debugInfo += ` - Wants Tables: ${queryIntent.wantsTables}\\n`;\n debugInfo += ` - Wants Media: ${queryIntent.wantsMedia}\\n`;\n debugInfo += ` - Countries: ${queryIntent.countries.join(\", \") || \"None\"}\\n`;\n debugInfo += ` - Companies: ${queryIntent.companies.join(\", \") || \"None\"}\\n\\n`;\n\n debugInfo += `Date Range Analysis:\\n`;\n debugInfo += ` - Start Date: ${dateRange.startDate}\\n`;\n debugInfo += ` - End Date: ${dateRange.endDate}\\n`;\n debugInfo += ` - Is Latest: ${dateRange.isLatest || false}\\n`;\n debugInfo += ` - Is Default: ${dateRange.isDefault || false}\\n`;\n debugInfo += ` - Searching for Media: ${dateRange.searchingForMedia || false}\\n\\n`;\n\n debugInfo += `Extracted Metrics: ${metrics.join(\", \") || \"None\"}\\n\\n`;\n\n // Build the filter that would be used\n let pineconeFilter = {};\n if (queryIntent.wantsCharts) pineconeFilter.has_charts = true;\n if (queryIntent.wantsMaps) pineconeFilter.has_maps = true;\n if (queryIntent.wantsTables) pineconeFilter.has_tables = true;\n if (queryIntent.countries.includes(\"Egypt\")) {\n pineconeFilter.primary_country = \"Eygpt\"; // Handle misspelling\n }\n\n debugInfo += `Pinecone Filter: ${JSON.stringify(pineconeFilter, null, 2)}\\n`;\n\n return new Response(\n JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n has_charts: meta.has_charts,\n has_maps: meta.has_maps,\n has_tables: meta.has_tables,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) =>\n new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime()\n );\n\n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n\n // Format the response with all metadata\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10 with full metadata:\\n`;\n\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n articleInfo += ` 🏷️ Has Media: Charts=${d.has_charts} Maps=${d.has_maps} Tables=${d.has_tables}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === \"string\") {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || \"N/A\"}: ${parsed.context || \"N/A\"}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === \"object\") {\n // If it's already an object\n articleInfo += ` • ${km.metric || \"N/A\"}: ${km.context || \"N/A\"}\\n`;\n }\n });\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return new Response(\n JSON.stringify({\n response: summary + formattedArticles,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n const queryIntent = analyzeQueryIntent(message);\n\n if (queryIntent.wantsMedia) {\n let mediaContext = \"\\n\\n[System Note: User is looking for \";\n const mediaTypes = [];\n if (queryIntent.wantsCharts) mediaTypes.push(\"charts\");\n if (queryIntent.wantsMaps) mediaTypes.push(\"maps\");\n if (queryIntent.wantsTables) mediaTypes.push(\"tables\");\n mediaContext += mediaTypes.join(\" and \");\n mediaContext +=\n \". Relevant media from search results is automatically displayed. Focus on describing what these visualizations show.]\";\n enhancedMessage += mediaContext;\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from the top search results are automatically displayed to the user in a side panel. Focus on analyzing the data and insights rather than listing media IDs.]\";\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500; // Limit text sent to OpenAI\n if (textContent.length > maxTextLength) {\n // Keep the beginning which usually has the most important info\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent, // Now truncated for faster processing\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n\n // If user is looking for specific media types, prioritize those\n const queryIntent = analyzeQueryIntent(parsedArgs.query);\n let resultsToScan = formattedResults.slice(0, 5); // Default: top 5\n\n if (queryIntent.wantsMedia) {\n // If looking for specific media, scan more results and filter\n resultsToScan = formattedResults.slice(0, 10);\n\n // Filter to only articles that have the requested media type\n if (queryIntent.wantsCharts) {\n resultsToScan = resultsToScan.filter(r => r.all_charts.length > 0);\n }\n if (queryIntent.wantsMaps) {\n resultsToScan = resultsToScan.filter(r => r.all_maps.length > 0);\n }\n if (queryIntent.wantsTables) {\n resultsToScan = resultsToScan.filter(r => r.all_tables.length > 0);\n }\n }\n\n resultsToScan.forEach((result, index) => {\n // Add charts if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsCharts) {\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n\n // Add maps if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsMaps) {\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n\n // Add tables if requested or by default (limit to first 2 per article)\n if (!queryIntent.wantsMedia || queryIntent.wantsTables) {\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",21,[142],{"_143":144,"_99":282,"_104":283,"_146":132,"_161":284},32,["D",1748615500293],[285],{"_102":286,"_146":132,"_143":144,"_95":142,"_154":269,"_152":153,"_99":287,"_145":-5,"_104":283,"_147":282,"_157":-5,"_148":27,"_156":27,"_98":288},"ccf2a5d6-3d62-11f0-ab42-76b3cceeab13",20,[142],{"_143":144,"_99":151,"_104":290,"_146":132,"_161":291},["D",1748614637890],[292],{"_102":293,"_146":132,"_143":144,"_95":142,"_154":294,"_152":153,"_99":295,"_145":-5,"_104":290,"_147":151,"_157":-5,"_148":27,"_156":27,"_98":296},"caea5ea2-3d60-11f0-ac65-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\")\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\")\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string, queryIntent: any = null) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n \n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true\n };\n }\n \n // Check for specific months\n const monthYearMatch = query.match(/\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i);\n if (monthYearMatch) {\n const monthNames = [\"january\", \"february\", \"march\", \"april\", \"may\", \"june\", \"july\", \"august\", \"september\", \"october\", \"november\", \"december\"];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true\n };\n }\n \n // Check for relative time indicators\n if (lowerQuery.includes('latest') || lowerQuery.includes('most recent') || lowerQuery.includes('newest') || lowerQuery.includes('current')) {\n // If searching for specific media (maps, charts, tables), use longer timeframe\n // since media items are less frequent than articles\n let monthsBack = 1; // Default for general queries\n \n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6; // Look back 6 months for media items\n }\n \n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n searchingForMedia: queryIntent && queryIntent.wantsMedia\n };\n }\n \n if (lowerQuery.includes('recent')) {\n // Last 3 months for \"recent\" (or 6 for media)\n let monthsBack = 3;\n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6;\n }\n \n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false\n };\n }\n \n if (lowerQuery.includes('this year')) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true\n };\n }\n \n if (lowerQuery.includes('last year')) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true\n };\n }\n \n // Default: Last 3 months for general queries, 6 months for media\n let monthsBack = 3;\n if (queryIntent && queryIntent.wantsMedia) {\n monthsBack = 6;\n }\n \n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - monthsBack);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true\n };\n}\n\n// Analyze query for specific search intents (media, companies, etc.)\nfunction analyzeQueryIntent(query: string) {\n const lowerQuery = query.toLowerCase();\n const intent = {\n wantsCharts: false,\n wantsMaps: false,\n wantsTables: false,\n wantsMedia: false,\n companies: [],\n countries: []\n };\n \n // Check for media requests\n if (lowerQuery.includes('chart') || lowerQuery.includes('graph') || lowerQuery.includes('visualiz')) {\n intent.wantsCharts = true;\n intent.wantsMedia = true;\n }\n \n if (lowerQuery.includes('map') || lowerQuery.includes('geography') || lowerQuery.includes('location')) {\n intent.wantsMaps = true;\n intent.wantsMedia = true;\n }\n \n if (lowerQuery.includes('table') || lowerQuery.includes('data table')) {\n intent.wantsTables = true;\n intent.wantsMedia = true;\n }\n \n // Check for country mentions\n const countries = ['egypt', 'saudi', 'uae', 'qatar', 'kuwait', 'iraq', 'iran', 'israel', 'jordan', 'lebanon', 'syria', 'turkey', 'cyprus'];\n countries.forEach(country => {\n if (lowerQuery.includes(country)) {\n // Capitalize first letter for proper matching\n let properCountry = country.charAt(0).toUpperCase() + country.slice(1);\n // Fix common misspellings in metadata\n if (properCountry === 'Egypt') {\n intent.countries.push('Egypt', 'Eygpt'); // Include misspelling found in data\n } else {\n intent.countries.push(properCountry);\n }\n }\n });\n \n // Check for company mentions (add more as needed)\n const companies = ['chevron', 'exxon', 'bp', 'shell', 'eni', 'total', 'apache', 'wintershall'];\n companies.forEach(company => {\n if (lowerQuery.includes(company)) {\n intent.companies.push(company);\n }\n });\n \n return intent;\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n \n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n \n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n \n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n \n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n \n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n \n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n \n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined\n });\n \n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n \n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n \n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n \n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n \n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n \n // Analyze query for temporal indicators\n const dateRange = analyzeQueryTemporal(query);\n console.log(\"Date range analysis:\", dateRange);\n \n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n \n // If we have specific metrics, try metric-based search first\n if (queryMetrics.length > 0) {\n console.log(\"Searching by metrics first...\");\n const metricResults = await searchByMetrics(queryMetrics, dateRange);\n if (metricResults.length > 0) {\n console.log(`Found ${metricResults.length} results matching metrics`);\n // If we found good metric matches, prioritize them\n // But also do a regular search to complement\n }\n }\n \n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n \n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone WITHOUT date range (we'll filter after)\n let pineconeFilter = {};\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n \n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n \n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n \n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n \n console.log(`${searchResults.matches.length} results after date filtering`);\n \n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n \n // If we got metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n \n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n \n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n \n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(' ');\n const [month, day, year] = datePart.split('/');\n const [hour, minute] = timePart ? timePart.split(':') : ['0', '0'];\n \n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n \n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n \n // Debug endpoint for testing specific queries\n if (message.startsWith(\"DEBUG_QUERY:\")) {\n const testQuery = message.replace(\"DEBUG_QUERY:\", \"\").trim();\n const queryIntent = analyzeQueryIntent(testQuery);\n const dateRange = analyzeQueryTemporal(testQuery, queryIntent);\n const metrics = extractMetricsFromQuery(testQuery);\n \n let debugInfo = `Query Analysis for: \"${testQuery}\"\\n\\n`;\n debugInfo += `Intent Analysis:\\n`;\n debugInfo += ` - Wants Charts: ${queryIntent.wantsCharts}\\n`;\n debugInfo += ` - Wants Maps: ${queryIntent.wantsMaps}\\n`;\n debugInfo += ` - Wants Tables: ${queryIntent.wantsTables}\\n`;\n debugInfo += ` - Wants Media: ${queryIntent.wantsMedia}\\n`;\n debugInfo += ` - Countries: ${queryIntent.countries.join(', ') || 'None'}\\n`;\n debugInfo += ` - Companies: ${queryIntent.companies.join(', ') || 'None'}\\n\\n`;\n \n debugInfo += `Date Range Analysis:\\n`;\n debugInfo += ` - Start Date: ${dateRange.startDate}\\n`;\n debugInfo += ` - End Date: ${dateRange.endDate}\\n`;\n debugInfo += ` - Is Latest: ${dateRange.isLatest || false}\\n`;\n debugInfo += ` - Is Default: ${dateRange.isDefault || false}\\n`;\n debugInfo += ` - Searching for Media: ${dateRange.searchingForMedia || false}\\n\\n`;\n \n debugInfo += `Extracted Metrics: ${metrics.join(', ') || 'None'}\\n\\n`;\n \n // Build the filter that would be used\n let pineconeFilter = {};\n if (queryIntent.wantsCharts) pineconeFilter.has_charts = true;\n if (queryIntent.wantsMaps) pineconeFilter.has_maps = true;\n if (queryIntent.wantsTables) pineconeFilter.has_tables = true;\n if (queryIntent.countries.includes('Egypt')) {\n pineconeFilter.primary_country = 'Eygpt'; // Handle misspelling\n }\n \n debugInfo += `Pinecone Filter: ${JSON.stringify(pineconeFilter, null, 2)}\\n`;\n \n return new Response(JSON.stringify({\n response: debugInfo,\n threadId: \"debug\",\n status: \"success\"\n }), { \n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\"\n }\n });\n }\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n \n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n \n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n has_charts: meta.has_charts,\n has_maps: meta.has_maps,\n has_tables: meta.has_tables\n };\n });\n \n // Sort by ISO date\n enrichedData.sort((a, b) => new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime());\n \n // Count articles with media\n const articlesWithMaps = enrichedData.filter(d => d.all_maps.length > 0).length;\n const articlesWithCharts = enrichedData.filter(d => d.all_charts.length > 0).length;\n const articlesWithTables = enrichedData.filter(d => d.all_tables.length > 0).length;\n \n // Format the response with all metadata\n let summary = `Found ${enrichedData.length} articles:\\n`;\n summary += `📊 Articles with charts: ${articlesWithCharts}\\n`;\n summary += `🗺️ Articles with maps: ${articlesWithMaps}\\n`;\n summary += `📋 Articles with tables: ${articlesWithTables}\\n\\n`;\n summary += `Showing latest 10 with full metadata:\\n`;\n \n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i+1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || 'missing'}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || 'N/A'} | Section: ${d.section || 'N/A'}\\n`;\n articleInfo += ` 🏷️ Has Media: Charts=${d.has_charts} Maps=${d.has_maps} Tables=${d.has_tables}\\n`;\n \n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(', ')}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(', ')}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(', ')}\\n`;\n }\n \n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(', ')}\\n`;\n }\n \n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === 'string') {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || 'N/A'}: ${parsed.context || 'N/A'}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === 'object') {\n // If it's already an object\n articleInfo += ` • ${km.metric || 'N/A'}: ${km.context || 'N/A'}\\n`;\n }\n });\n }\n \n return articleInfo;\n }).join('\\n' + '-'.repeat(60) + '\\n');\n \n return new Response(JSON.stringify({\n response: summary + formattedArticles,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\"\n }), { \n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\"\n }\n });\n }\n \n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n \n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n \n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\" // This is the critical header for v2 API\n };\n \n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n \n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({})\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n const queryIntent = analyzeQueryIntent(message);\n \n if (queryIntent.wantsMedia) {\n let mediaContext = \"\\n\\n[System Note: User is looking for \";\n const mediaTypes = [];\n if (queryIntent.wantsCharts) mediaTypes.push(\"charts\");\n if (queryIntent.wantsMaps) mediaTypes.push(\"maps\");\n if (queryIntent.wantsTables) mediaTypes.push(\"tables\");\n mediaContext += mediaTypes.join(\" and \");\n mediaContext += \". Relevant media from search results is automatically displayed. Focus on describing what these visualizations show.]\";\n enhancedMessage += mediaContext;\n } else {\n enhancedMessage += \"\\n\\n[System Note: Charts, maps, and tables from the top search results are automatically displayed to the user in a side panel. Focus on analyzing the data and insights rather than listing media IDs.]\";\n }\n \n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage\n })\n });\n \n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID\n })\n });\n \n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n \n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n \n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n \n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n \n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n \n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500; // Limit text sent to OpenAI\n if (textContent.length > maxTextLength) {\n // Keep the beginning which usually has the most important info\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent, // Now truncated for faster processing\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country\n };\n });\n \n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n \n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n \n // If user is looking for specific media types, prioritize those\n const queryIntent = analyzeQueryIntent(parsedArgs.query);\n let resultsToScan = formattedResults.slice(0, 5); // Default: top 5\n \n if (queryIntent.wantsMedia) {\n // If looking for specific media, scan more results and filter\n resultsToScan = formattedResults.slice(0, 10);\n \n // Filter to only articles that have the requested media type\n if (queryIntent.wantsCharts) {\n resultsToScan = resultsToScan.filter(r => r.all_charts.length > 0);\n }\n if (queryIntent.wantsMaps) {\n resultsToScan = resultsToScan.filter(r => r.all_maps.length > 0);\n }\n if (queryIntent.wantsTables) {\n resultsToScan = resultsToScan.filter(r => r.all_tables.length > 0);\n }\n }\n \n resultsToScan.forEach((result, index) => {\n // Add charts if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsCharts) {\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n }\n \n // Add maps if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsMaps) {\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n }\n \n // Add tables if requested or by default (limit to first 2 per article)\n if (!queryIntent.wantsMedia || queryIntent.wantsTables) {\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`\n });\n }\n });\n }\n });\n \n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n \n console.log(\"Results after sorting:\", formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score\n })));\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n \n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n \n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n \n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders } // v2 headers\n );\n \n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n \n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n \n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n \n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs })\n }\n );\n \n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders } // v2 headers\n );\n \n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n \n const messages = await messagesResponse.json();\n \n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n \n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n \n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n \n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n \n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",19,[142],{"_143":144,"_99":200,"_104":298,"_146":132,"_161":299},["D",1748614160104],[300],{"_102":301,"_146":132,"_143":144,"_95":142,"_154":302,"_152":153,"_99":303,"_145":-5,"_104":298,"_147":200,"_157":-5,"_148":27,"_156":27,"_98":304},"ae226400-3d5f-11f0-a851-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n\n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for specific months\n const monthYearMatch = query.match(\n /\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i,\n );\n if (monthYearMatch) {\n const monthNames = [\n \"january\",\n \"february\",\n \"march\",\n \"april\",\n \"may\",\n \"june\",\n \"july\",\n \"august\",\n \"september\",\n \"october\",\n \"november\",\n \"december\",\n ];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for relative time indicators\n if (\n lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n ) {\n // Last 1 month for \"latest\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 1);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n };\n }\n\n if (lowerQuery.includes(\"recent\")) {\n // Last 3 months for \"recent\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n };\n }\n\n if (lowerQuery.includes(\"this year\")) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n if (lowerQuery.includes(\"last year\")) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Default: Last 3 months for general queries\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true,\n };\n}\n\n// Analyze query for specific search intents (media, companies, etc.)\nfunction analyzeQueryIntent(query: string) {\n const lowerQuery = query.toLowerCase();\n const intent = {\n wantsCharts: false,\n wantsMaps: false,\n wantsTables: false,\n wantsMedia: false,\n companies: [],\n countries: [],\n };\n\n // Check for media requests\n if (lowerQuery.includes(\"chart\") || lowerQuery.includes(\"graph\") || lowerQuery.includes(\"visualiz\")) {\n intent.wantsCharts = true;\n intent.wantsMedia = true;\n }\n\n if (lowerQuery.includes(\"map\") || lowerQuery.includes(\"geography\") || lowerQuery.includes(\"location\")) {\n intent.wantsMaps = true;\n intent.wantsMedia = true;\n }\n\n if (lowerQuery.includes(\"table\") || lowerQuery.includes(\"data table\")) {\n intent.wantsTables = true;\n intent.wantsMedia = true;\n }\n\n // Check for country mentions\n const countries = [\n \"egypt\",\n \"saudi\",\n \"uae\",\n \"qatar\",\n \"kuwait\",\n \"iraq\",\n \"iran\",\n \"israel\",\n \"jordan\",\n \"lebanon\",\n \"syria\",\n \"turkey\",\n \"cyprus\",\n ];\n countries.forEach(country => {\n if (lowerQuery.includes(country)) {\n // Capitalize first letter for proper matching\n let properCountry = country.charAt(0).toUpperCase() + country.slice(1);\n // Fix common misspellings in metadata\n if (properCountry === \"Egypt\") {\n intent.countries.push(\"Egypt\", \"Eygpt\"); // Include misspelling found in data\n } else {\n intent.countries.push(properCountry);\n }\n }\n });\n\n // Check for company mentions (add more as needed)\n const companies = [\"chevron\", \"exxon\", \"bp\", \"shell\", \"eni\", \"total\", \"apache\", \"wintershall\"];\n companies.forEach(company => {\n if (lowerQuery.includes(company)) {\n intent.companies.push(company);\n }\n });\n\n return intent;\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n\n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n\n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n\n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n\n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n\n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n\n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n\n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined,\n });\n\n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n\n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n\n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Analyze query for temporal indicators\n const dateRange = analyzeQueryTemporal(query);\n console.log(\"Date range analysis:\", dateRange);\n\n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n\n // If we have specific metrics, try metric-based search first\n if (queryMetrics.length > 0) {\n console.log(\"Searching by metrics first...\");\n const metricResults = await searchByMetrics(queryMetrics, dateRange);\n if (metricResults.length > 0) {\n console.log(`Found ${metricResults.length} results matching metrics`);\n // If we found good metric matches, prioritize them\n // But also do a regular search to complement\n }\n }\n\n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone WITHOUT date range (we'll filter after)\n let pineconeFilter = {};\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n\n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n console.log(`${searchResults.matches.length} results after date filtering`);\n\n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n\n // If we got metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n\n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) =>\n new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime()\n );\n\n // Format the response with all metadata\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === \"string\") {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || \"N/A\"}: ${parsed.context || \"N/A\"}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === \"object\") {\n // If it's already an object\n articleInfo += ` • ${km.metric || \"N/A\"}: ${km.context || \"N/A\"}\\n`;\n }\n });\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return new Response(\n JSON.stringify({\n response:\n `Found ${enrichedData.length} articles. Showing latest 10 with full metadata:\\n${formattedArticles}`,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n const queryIntent = analyzeQueryIntent(message);\n\n if (queryIntent.wantsMedia) {\n let mediaContext = \"\\n\\n[System Note: User is looking for \";\n const mediaTypes = [];\n if (queryIntent.wantsCharts) mediaTypes.push(\"charts\");\n if (queryIntent.wantsMaps) mediaTypes.push(\"maps\");\n if (queryIntent.wantsTables) mediaTypes.push(\"tables\");\n mediaContext += mediaTypes.join(\" and \");\n mediaContext +=\n \". Relevant media from search results is automatically displayed. Focus on describing what these visualizations show.]\";\n enhancedMessage += mediaContext;\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from the top search results are automatically displayed to the user in a side panel. Focus on analyzing the data and insights rather than listing media IDs.]\";\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500; // Limit text sent to OpenAI\n if (textContent.length > maxTextLength) {\n // Keep the beginning which usually has the most important info\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent, // Now truncated for faster processing\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n\n // If user is looking for specific media types, prioritize those\n const queryIntent = analyzeQueryIntent(parsedArgs.query);\n let resultsToScan = formattedResults.slice(0, 5); // Default: top 5\n\n if (queryIntent.wantsMedia) {\n // If looking for specific media, scan more results and filter\n resultsToScan = formattedResults.slice(0, 10);\n\n // Filter to only articles that have the requested media type\n if (queryIntent.wantsCharts) {\n resultsToScan = resultsToScan.filter(r => r.all_charts.length > 0);\n }\n if (queryIntent.wantsMaps) {\n resultsToScan = resultsToScan.filter(r => r.all_maps.length > 0);\n }\n if (queryIntent.wantsTables) {\n resultsToScan = resultsToScan.filter(r => r.all_tables.length > 0);\n }\n }\n\n resultsToScan.forEach((result, index) => {\n // Add charts if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsCharts) {\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n\n // Add maps if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsMaps) {\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n\n // Add tables if requested or by default (limit to first 2 per article)\n if (!queryIntent.wantsMedia || queryIntent.wantsTables) {\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",18,[142],{"_143":144,"_99":209,"_104":306,"_146":132,"_161":307},["D",1748613917003],[308],{"_102":309,"_146":132,"_143":144,"_95":142,"_154":310,"_152":153,"_99":311,"_145":-5,"_104":306,"_147":209,"_157":-5,"_148":27,"_156":27,"_98":312},"1d3bd106-3d5f-11f0-bf8f-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n\n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for specific months\n const monthYearMatch = query.match(\n /\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i,\n );\n if (monthYearMatch) {\n const monthNames = [\n \"january\",\n \"february\",\n \"march\",\n \"april\",\n \"may\",\n \"june\",\n \"july\",\n \"august\",\n \"september\",\n \"october\",\n \"november\",\n \"december\",\n ];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for relative time indicators\n if (\n lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n ) {\n // Last 1 month for \"latest\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 1);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n };\n }\n\n if (lowerQuery.includes(\"recent\")) {\n // Last 3 months for \"recent\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n };\n }\n\n if (lowerQuery.includes(\"this year\")) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n if (lowerQuery.includes(\"last year\")) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Default: Last 3 months for general queries\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true,\n };\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n\n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n\n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n\n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n\n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n\n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n\n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n\n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined,\n });\n\n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n\n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n\n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Analyze query for temporal indicators\n const dateRange = analyzeQueryTemporal(query);\n console.log(\"Date range analysis:\", dateRange);\n\n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n\n // If we have specific metrics, try metric-based search first\n if (queryMetrics.length > 0) {\n console.log(\"Searching by metrics first...\");\n const metricResults = await searchByMetrics(queryMetrics, dateRange);\n if (metricResults.length > 0) {\n console.log(`Found ${metricResults.length} results matching metrics`);\n // If we found good metric matches, prioritize them\n // But also do a regular search to complement\n }\n }\n\n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone WITHOUT date range (we'll filter after)\n let pineconeFilter = {};\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n\n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n console.log(`${searchResults.matches.length} results after date filtering`);\n\n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n\n // If we got metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n\n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) =>\n new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime()\n );\n\n // Format the response with all metadata\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === \"string\") {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || \"N/A\"}: ${parsed.context || \"N/A\"}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === \"object\") {\n // If it's already an object\n articleInfo += ` • ${km.metric || \"N/A\"}: ${km.context || \"N/A\"}\\n`;\n }\n });\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return new Response(\n JSON.stringify({\n response:\n `Found ${enrichedData.length} articles. Showing latest 10 with full metadata:\\n${formattedArticles}`,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Analyze the message to add context for the assistant\n let enhancedMessage = message;\n const queryIntent = analyzeQueryIntent(message);\n\n if (queryIntent.wantsMedia) {\n let mediaContext = \"\\n\\n[System Note: User is looking for \";\n const mediaTypes = [];\n if (queryIntent.wantsCharts) mediaTypes.push(\"charts\");\n if (queryIntent.wantsMaps) mediaTypes.push(\"maps\");\n if (queryIntent.wantsTables) mediaTypes.push(\"tables\");\n mediaContext += mediaTypes.join(\" and \");\n mediaContext +=\n \". Relevant media from search results is automatically displayed. Focus on describing what these visualizations show.]\";\n enhancedMessage += mediaContext;\n } else {\n enhancedMessage +=\n \"\\n\\n[System Note: Charts, maps, and tables from the top search results are automatically displayed to the user in a side panel. Focus on analyzing the data and insights rather than listing media IDs.]\";\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: enhancedMessage,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500; // Limit text sent to OpenAI\n if (textContent.length > maxTextLength) {\n // Keep the beginning which usually has the most important info\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent, // Now truncated for faster processing\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n\n // If user is looking for specific media types, prioritize those\n const queryIntent = analyzeQueryIntent(parsedArgs.query);\n let resultsToScan = formattedResults.slice(0, 5); // Default: top 5\n\n if (queryIntent.wantsMedia) {\n // If looking for specific media, scan more results and filter\n resultsToScan = formattedResults.slice(0, 10);\n\n // Filter to only articles that have the requested media type\n if (queryIntent.wantsCharts) {\n resultsToScan = resultsToScan.filter(r => r.all_charts.length > 0);\n }\n if (queryIntent.wantsMaps) {\n resultsToScan = resultsToScan.filter(r => r.all_maps.length > 0);\n }\n if (queryIntent.wantsTables) {\n resultsToScan = resultsToScan.filter(r => r.all_tables.length > 0);\n }\n }\n\n resultsToScan.forEach((result, index) => {\n // Add charts if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsCharts) {\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n\n // Add maps if requested or by default\n if (!queryIntent.wantsMedia || queryIntent.wantsMaps) {\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n\n // Add tables if requested or by default (limit to first 2 per article)\n if (!queryIntent.wantsMedia || queryIntent.wantsTables) {\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From: ${result.title.substring(0, 50)}... (${result.date})`,\n });\n }\n });\n }\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",17,[142],{"_143":144,"_99":218,"_104":314,"_146":132,"_161":315},["D",1748613368272],[316],{"_102":317,"_146":132,"_143":144,"_95":142,"_154":318,"_152":153,"_99":319,"_145":-5,"_104":314,"_147":218,"_157":-5,"_148":27,"_156":27,"_98":320},"d62a53a6-3d5d-11f0-8eab-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n\n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for specific months\n const monthYearMatch = query.match(\n /\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i,\n );\n if (monthYearMatch) {\n const monthNames = [\n \"january\",\n \"february\",\n \"march\",\n \"april\",\n \"may\",\n \"june\",\n \"july\",\n \"august\",\n \"september\",\n \"october\",\n \"november\",\n \"december\",\n ];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for relative time indicators\n if (\n lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n ) {\n // Last 1 month for \"latest\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 1);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n };\n }\n\n if (lowerQuery.includes(\"recent\")) {\n // Last 3 months for \"recent\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n };\n }\n\n if (lowerQuery.includes(\"this year\")) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n if (lowerQuery.includes(\"last year\")) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Default: Last 3 months for general queries\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true,\n };\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n\n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n\n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n\n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n\n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n\n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n\n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n\n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined,\n });\n\n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n\n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n\n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Analyze query for temporal indicators\n const dateRange = analyzeQueryTemporal(query);\n console.log(\"Date range analysis:\", dateRange);\n\n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n\n // If we have specific metrics, try metric-based search first\n if (queryMetrics.length > 0) {\n console.log(\"Searching by metrics first...\");\n const metricResults = await searchByMetrics(queryMetrics, dateRange);\n if (metricResults.length > 0) {\n console.log(`Found ${metricResults.length} results matching metrics`);\n // If we found good metric matches, prioritize them\n // But also do a regular search to complement\n }\n }\n\n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone WITHOUT date range (we'll filter after)\n let pineconeFilter = {};\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n\n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n console.log(`${searchResults.matches.length} results after date filtering`);\n\n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n\n // If we got metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n\n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) =>\n new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime()\n );\n\n // Format the response with all metadata\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n // Handle both string and object formats\n if (typeof km === \"string\") {\n // If it's a string, try to parse it\n try {\n const parsed = JSON.parse(km);\n articleInfo += ` • ${parsed.metric || \"N/A\"}: ${parsed.context || \"N/A\"}\\n`;\n } catch {\n articleInfo += ` • ${km}\\n`;\n }\n } else if (km && typeof km === \"object\") {\n // If it's already an object\n articleInfo += ` • ${km.metric || \"N/A\"}: ${km.context || \"N/A\"}\\n`;\n }\n });\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return new Response(\n JSON.stringify({\n response:\n `Found ${enrichedData.length} articles. Showing latest 10 with full metadata:\\n${formattedArticles}`,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message\n + \"\\n\\n[System Note: Charts, maps, and tables from the top search results are automatically displayed to the user in a side panel. Focus on analyzing the data and insights rather than listing media IDs.]\",\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // OPTIMIZATION: Only send text excerpts to OpenAI\n let textContent = meta.text || \"\";\n const maxTextLength = 500; // Limit text sent to OpenAI\n if (textContent.length > maxTextLength) {\n // Keep the beginning which usually has the most important info\n textContent = textContent.substring(0, maxTextLength) + \"... [truncated]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent, // Now truncated for faster processing\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n // AUTO-DISPLAY MEDIA: Collect all media from top results\n const autoDisplayMedia = [];\n const topResultsForMedia = formattedResults.slice(0, 3); // Top 3 most recent articles\n\n topResultsForMedia.forEach((result, index) => {\n // Add charts\n result.all_charts.forEach(chartId => {\n if (chartId && !autoDisplayMedia.find(m => m.id === chartId)) {\n autoDisplayMedia.push({\n id: chartId,\n type: \"chart\",\n title: MEDIA_TITLES[chartId] || `Chart ${chartId}`,\n article_id: result.article_id,\n relevance: `From article ${index + 1}: ${result.title.substring(0, 40)}...`,\n });\n }\n });\n\n // Add maps\n result.all_maps.forEach(mapId => {\n if (mapId && !autoDisplayMedia.find(m => m.id === mapId)) {\n autoDisplayMedia.push({\n id: mapId,\n type: \"map\",\n title: MEDIA_TITLES[mapId] || `Map ${mapId}`,\n article_id: result.article_id,\n relevance: `From article ${index + 1}: ${result.title.substring(0, 40)}...`,\n });\n }\n });\n\n // Add tables (limit to first 2 per article)\n result.all_tables.slice(0, 2).forEach(tableId => {\n if (tableId && !autoDisplayMedia.find(m => m.id === tableId)) {\n autoDisplayMedia.push({\n id: tableId,\n type: \"table\",\n title: MEDIA_TITLES[tableId] || `Table ${tableId}`,\n article_id: result.article_id,\n relevance: `From article ${index + 1}: ${result.title.substring(0, 40)}...`,\n });\n }\n });\n });\n\n // Add auto-display media to the media displays\n if (autoDisplayMedia.length > 0) {\n console.log(`Auto-displaying ${autoDisplayMedia.length} media items from search results`);\n mediaDisplays.push(...autoDisplayMedia);\n }\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",16,[142],{"_143":144,"_99":226,"_104":322,"_146":132,"_161":323},["D",1748613016000],[324],{"_102":325,"_146":132,"_143":144,"_95":142,"_154":326,"_152":153,"_99":327,"_145":-5,"_104":322,"_147":226,"_157":-5,"_148":27,"_156":27,"_98":328},"04317eec-3d5d-11f0-bb69-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n\n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for specific months\n const monthYearMatch = query.match(\n /\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i,\n );\n if (monthYearMatch) {\n const monthNames = [\n \"january\",\n \"february\",\n \"march\",\n \"april\",\n \"may\",\n \"june\",\n \"july\",\n \"august\",\n \"september\",\n \"october\",\n \"november\",\n \"december\",\n ];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for relative time indicators\n if (\n lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n ) {\n // Last 1 month for \"latest\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 1);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n };\n }\n\n if (lowerQuery.includes(\"recent\")) {\n // Last 3 months for \"recent\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n };\n }\n\n if (lowerQuery.includes(\"this year\")) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n if (lowerQuery.includes(\"last year\")) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Default: Last 3 months for general queries\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true,\n };\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n\n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n\n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n\n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n\n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n\n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n\n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n\n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined,\n });\n\n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n\n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n\n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Analyze query for temporal indicators\n const dateRange = analyzeQueryTemporal(query);\n console.log(\"Date range analysis:\", dateRange);\n\n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n\n // If we have specific metrics, try metric-based search first\n if (queryMetrics.length > 0) {\n console.log(\"Searching by metrics first...\");\n const metricResults = await searchByMetrics(queryMetrics, dateRange);\n if (metricResults.length > 0) {\n console.log(`Found ${metricResults.length} results matching metrics`);\n // If we found good metric matches, prioritize them\n // But also do a regular search to complement\n }\n }\n\n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone WITHOUT date range (we'll filter after)\n let pineconeFilter = {};\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n\n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n console.log(`${searchResults.matches.length} results after date filtering`);\n\n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n\n // If we got metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n\n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) =>\n new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime()\n );\n\n // Format the response with all metadata\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n**[${i + 1}] ${d.title}**\\n`;\n articleInfo += ` 📅 Date: ${d.date} | ISO: ${d.date_iso || \"missing\"}\\n`;\n articleInfo += ` 🆔 Article: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += ` 🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n\n // Media items\n if (d.all_charts.length > 0) {\n articleInfo += ` 📊 Charts (${d.all_charts.length}): ${d.all_charts.join(\", \")}\\n`;\n }\n if (d.all_maps.length > 0) {\n articleInfo += ` 🗺️ Maps (${d.all_maps.length}): ${d.all_maps.join(\", \")}\\n`;\n }\n if (d.all_tables.length > 0) {\n articleInfo += ` 📋 Tables (${d.all_tables.length}): ${d.all_tables.join(\", \")}\\n`;\n }\n\n // Companies\n if (d.companies.length > 0) {\n articleInfo += ` 🏢 Companies (${d.companies.length}): ${d.companies.join(\", \")}\\n`;\n }\n\n // Key Metrics\n if (d.key_metrics.length > 0) {\n articleInfo += ` 💰 Key Metrics (${d.key_metrics.length}):\\n`;\n d.key_metrics.forEach(km => {\n articleInfo += ` • ${km.metric}: ${km.context}\\n`;\n });\n }\n\n return articleInfo;\n }).join(\"\\n\" + \"-\".repeat(60) + \"\\n\");\n\n return new Response(\n JSON.stringify({\n response:\n `Found ${enrichedData.length} articles. Showing latest 10 with full metadata:\\n${formattedArticles}`,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Don't truncate text - the assistant needs full context\n let textContent = meta.text || \"\";\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",15,[142],{"_143":144,"_99":235,"_104":330,"_146":132,"_161":331},["D",1748612777784],[332],{"_102":333,"_146":132,"_143":144,"_95":142,"_154":334,"_152":153,"_99":335,"_145":-5,"_104":330,"_147":235,"_157":-5,"_148":27,"_156":27,"_98":336},"7634e82c-3d5c-11f0-9084-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n\n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for specific months\n const monthYearMatch = query.match(\n /\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i,\n );\n if (monthYearMatch) {\n const monthNames = [\n \"january\",\n \"february\",\n \"march\",\n \"april\",\n \"may\",\n \"june\",\n \"july\",\n \"august\",\n \"september\",\n \"october\",\n \"november\",\n \"december\",\n ];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for relative time indicators\n if (\n lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n ) {\n // Last 1 month for \"latest\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 1);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n };\n }\n\n if (lowerQuery.includes(\"recent\")) {\n // Last 3 months for \"recent\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n };\n }\n\n if (lowerQuery.includes(\"this year\")) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n if (lowerQuery.includes(\"last year\")) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Default: Last 3 months for general queries\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true,\n };\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n\n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n\n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n\n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n\n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n\n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n\n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n\n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined,\n });\n\n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n\n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n\n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Analyze query for temporal indicators\n const dateRange = analyzeQueryTemporal(query);\n console.log(\"Date range analysis:\", dateRange);\n\n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n\n // If we have specific metrics, try metric-based search first\n if (queryMetrics.length > 0) {\n console.log(\"Searching by metrics first...\");\n const metricResults = await searchByMetrics(queryMetrics, dateRange);\n if (metricResults.length > 0) {\n console.log(`Found ${metricResults.length} results matching metrics`);\n // If we found good metric matches, prioritize them\n // But also do a regular search to complement\n }\n }\n\n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone WITHOUT date range (we'll filter after)\n let pineconeFilter = {};\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n\n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n console.log(`${searchResults.matches.length} results after date filtering`);\n\n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n\n // If we got metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n\n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const enrichedData = results.map(r => {\n const meta = r.metadata;\n\n // Parse JSON fields\n const all_charts = JSON.parse(meta.all_charts || \"[]\");\n const all_maps = JSON.parse(meta.all_maps || \"[]\");\n const all_tables = JSON.parse(meta.all_tables || \"[]\");\n const companies = JSON.parse(meta.companies || \"[]\");\n const key_metrics = JSON.parse(meta.key_metrics || \"[]\");\n\n return {\n date: meta.publish_date,\n date_iso: meta.publish_date_iso,\n parsed: parseMEESDate(meta.publish_date).toISOString(),\n title: meta.title ? meta.title.substring(0, 80) : \"No title\",\n article_id: meta.article_id,\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n all_charts,\n all_maps,\n all_tables,\n companies,\n key_metrics,\n primary_country: meta.primary_country,\n section: meta.section,\n };\n });\n\n // Sort by ISO date\n enrichedData.sort((a, b) =>\n new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime()\n );\n\n // Format the response with all metadata\n const formattedArticles = enrichedData.slice(0, 10).map((d, i) => {\n let articleInfo = `\\n${\"=\".repeat(80)}\\n`;\n articleInfo += `${i + 1}. ${d.title}\\n`;\n articleInfo += `${\"=\".repeat(80)}\\n`;\n articleInfo += `📅 Date: ${d.date} (ISO: ${d.date_iso || \"missing\"})\\n`;\n articleInfo += `🆔 Article ID: ${d.article_id} | Chunk: ${d.chunk_num}/${d.total_chunks}\\n`;\n articleInfo += `🌍 Country: ${d.primary_country || \"N/A\"} | Section: ${d.section || \"N/A\"}\\n`;\n\n // Media items\n articleInfo += `\\n📊 Charts (${d.all_charts.length}): ${\n d.all_charts.length > 0 ? d.all_charts.join(\", \") : \"None\"\n }\\n`;\n articleInfo += `🗺️ Maps (${d.all_maps.length}): ${d.all_maps.length > 0 ? d.all_maps.join(\", \") : \"None\"}\\n`;\n articleInfo += `📋 Tables (${d.all_tables.length}): ${\n d.all_tables.length > 0 ? d.all_tables.join(\", \") : \"None\"\n }\\n`;\n\n // Companies\n articleInfo += `\\n🏢 Companies (${d.companies.length}): ${\n d.companies.length > 0 ? d.companies.join(\", \") : \"None\"\n }\\n`;\n\n // Key Metrics Table\n if (d.key_metrics.length > 0) {\n articleInfo += `\\n💰 Key Metrics (${d.key_metrics.length}):\\n`;\n articleInfo += `┌${\"─\".repeat(20)}┬${\"─\".repeat(57)}┐\\n`;\n articleInfo += `│ ${\"Metric\".padEnd(18)} │ ${\"Context\".padEnd(55)} │\\n`;\n articleInfo += `├${\"─\".repeat(20)}┼${\"─\".repeat(57)}┤\\n`;\n\n d.key_metrics.forEach(km => {\n const metric = (km.metric || \"\").substring(0, 18).padEnd(18);\n const context = (km.context || \"\").substring(0, 55).padEnd(55);\n articleInfo += `│ ${metric} │ ${context} │\\n`;\n });\n\n articleInfo += `└${\"─\".repeat(20)}┴${\"─\".repeat(57)}┘\\n`;\n } else {\n articleInfo += `\\n💰 Key Metrics: None\\n`;\n }\n\n return articleInfo;\n }).join(\"\\n\");\n\n return new Response(\n JSON.stringify({\n response:\n `Found ${enrichedData.length} articles. Showing latest 10 with full metadata:\\n${formattedArticles}`,\n debug_data: enrichedData.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Don't truncate text - the assistant needs full context\n let textContent = meta.text || \"\";\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",14,[142],{"_143":144,"_99":243,"_104":338,"_146":132,"_161":339},["D",1748612326182],[340],{"_102":341,"_146":132,"_143":144,"_95":142,"_154":342,"_152":153,"_99":343,"_145":-5,"_104":338,"_147":243,"_157":-5,"_148":27,"_156":27,"_98":344},"6907fb4a-3d5b-11f0-96e0-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n\n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for specific months\n const monthYearMatch = query.match(\n /\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i,\n );\n if (monthYearMatch) {\n const monthNames = [\n \"january\",\n \"february\",\n \"march\",\n \"april\",\n \"may\",\n \"june\",\n \"july\",\n \"august\",\n \"september\",\n \"october\",\n \"november\",\n \"december\",\n ];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for relative time indicators\n if (\n lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n ) {\n // Last 1 month for \"latest\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 1);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n };\n }\n\n if (lowerQuery.includes(\"recent\")) {\n // Last 3 months for \"recent\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n };\n }\n\n if (lowerQuery.includes(\"this year\")) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n if (lowerQuery.includes(\"last year\")) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Default: Last 3 months for general queries\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true,\n };\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n\n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n\n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n\n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n\n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n\n // Build filter for metrics search WITHOUT date filter\n const filter = {};\n\n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n\n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 200, // Get more since we'll filter by date after\n includeMetadata: true,\n filter: Object.keys(filter).length > 0 ? filter : undefined,\n });\n\n // Filter by date range\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n const dateFilteredResults = results.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n // Filter results by matching metrics\n const matchedResults = dateFilteredResults.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n\n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n\n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Analyze query for temporal indicators\n const dateRange = analyzeQueryTemporal(query);\n console.log(\"Date range analysis:\", dateRange);\n\n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n\n // If we have specific metrics, try metric-based search first\n if (queryMetrics.length > 0) {\n console.log(\"Searching by metrics first...\");\n const metricResults = await searchByMetrics(queryMetrics, dateRange);\n if (metricResults.length > 0) {\n console.log(`Found ${metricResults.length} results matching metrics`);\n // If we found good metric matches, prioritize them\n // But also do a regular search to complement\n }\n }\n\n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone WITHOUT date range (we'll filter after)\n let pineconeFilter = {};\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone WITHOUT date filter (we'll filter after retrieval)\n // Get more results than needed to account for post-filtering\n const searchTopK = Math.min(topK * 3, 100); // Get 3x more to ensure enough after date filtering\n\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: searchTopK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results before date filtering`);\n\n // Filter results by date range AFTER retrieval\n const startDate = new Date(dateRange.startDate);\n const endDate = new Date(dateRange.endDate);\n\n searchResults.matches = searchResults.matches.filter(match => {\n const articleDate = new Date(match.metadata.publish_date_iso || match.metadata.publish_date);\n return articleDate >= startDate && articleDate <= endDate;\n });\n\n console.log(`${searchResults.matches.length} results after date filtering`);\n\n // Trim to requested topK after filtering\n searchResults.matches = searchResults.matches.slice(0, topK);\n\n // If we got metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n\n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const dates = results.map(r => ({\n date: r.metadata.publish_date,\n date_iso: r.metadata.publish_date_iso,\n parsed: parseMEESDate(r.metadata.publish_date).toISOString(),\n title: r.metadata.title ? r.metadata.title.substring(0, 60) : \"No title\",\n article_id: r.metadata.article_id,\n }));\n\n // Sort by ISO date\n dates.sort((a, b) => new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime());\n\n return new Response(\n JSON.stringify({\n response: `Found ${dates.length} articles. Latest 10:\\n\\n`\n + dates.slice(0, 10).map((d, i) =>\n `${i + 1}. ${d.title}\\n Date: ${d.date} (ISO: ${d.date_iso || \"missing\"})\\n ID: ${d.article_id}`\n ).join(\"\\n\\n\"),\n debug_data: dates.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Don't truncate text - the assistant needs full context\n let textContent = meta.text || \"\";\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",13,[142],{"_143":144,"_99":252,"_104":346,"_146":132,"_161":347},["D",1748612079603],[348],{"_102":349,"_146":132,"_143":144,"_95":142,"_154":350,"_152":153,"_99":351,"_145":-5,"_104":346,"_147":252,"_157":-5,"_148":27,"_156":27,"_98":352},"d60f2084-3d5a-11f0-937f-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\n// Analyze query for temporal indicators and return date range\nfunction analyzeQueryTemporal(query: string) {\n const lowerQuery = query.toLowerCase();\n const now = new Date();\n\n // Check for specific years (2020-2029)\n const yearMatch = query.match(/\\b(202[0-9])\\b/);\n if (yearMatch) {\n const year = parseInt(yearMatch[1]);\n return {\n startDate: new Date(year, 0, 1).toISOString(),\n endDate: new Date(year, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for specific months\n const monthYearMatch = query.match(\n /\\b(january|february|march|april|may|june|july|august|september|october|november|december)\\s+(\\d{4})\\b/i,\n );\n if (monthYearMatch) {\n const monthNames = [\n \"january\",\n \"february\",\n \"march\",\n \"april\",\n \"may\",\n \"june\",\n \"july\",\n \"august\",\n \"september\",\n \"october\",\n \"november\",\n \"december\",\n ];\n const monthIndex = monthNames.indexOf(monthYearMatch[1].toLowerCase());\n const year = parseInt(monthYearMatch[2]);\n const startDate = new Date(year, monthIndex, 1);\n const endDate = new Date(year, monthIndex + 1, 0, 23, 59, 59);\n return {\n startDate: startDate.toISOString(),\n endDate: endDate.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Check for relative time indicators\n if (\n lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n ) {\n // Last 1 month for \"latest\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 1);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isLatest: true,\n };\n }\n\n if (lowerQuery.includes(\"recent\")) {\n // Last 3 months for \"recent\"\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n };\n }\n\n if (lowerQuery.includes(\"this year\")) {\n return {\n startDate: new Date(now.getFullYear(), 0, 1).toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n if (lowerQuery.includes(\"last year\")) {\n const lastYear = now.getFullYear() - 1;\n return {\n startDate: new Date(lastYear, 0, 1).toISOString(),\n endDate: new Date(lastYear, 11, 31, 23, 59, 59).toISOString(),\n isSpecificPeriod: true,\n };\n }\n\n // Default: Last 3 months for general queries\n const startDate = new Date(now);\n startDate.setMonth(now.getMonth() - 3);\n return {\n startDate: startDate.toISOString(),\n endDate: now.toISOString(),\n isSpecificPeriod: false,\n isDefault: true,\n };\n}\n\n// Extract metrics/numbers from query\nfunction extractMetricsFromQuery(query: string) {\n const metrics = [];\n\n // Extract currency amounts\n const currencyMatches = query.match(/\\$[\\d,]+\\.?\\d*[bm]?n?\\b/gi);\n if (currencyMatches) {\n metrics.push(...currencyMatches.map(m => m.toLowerCase()));\n }\n\n // Extract percentages\n const percentMatches = query.match(/\\d+\\.?\\d*%/g);\n if (percentMatches) {\n metrics.push(...percentMatches);\n }\n\n // Extract large numbers with units\n const numberMatches = query.match(/\\d+\\.?\\d*\\s*(bcm|mcm|b\\/d|kb\\/d|mb\\/d|mw|gw|twh|mtpa)/gi);\n if (numberMatches) {\n metrics.push(...numberMatches.map(m => m.toLowerCase()));\n }\n\n return metrics;\n}\n\n// Search for articles by metrics\nasync function searchByMetrics(metrics: string[], dateRange: any) {\n const index = pinecone.index(\"mees-articles\");\n\n // Build filter for metrics search\n const filter = {\n publish_date_iso: {\n $gte: dateRange.startDate,\n $lte: dateRange.endDate,\n },\n };\n\n // Get a sample embedding (we need one for Pinecone query)\n const dummyEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: \"energy metrics data\",\n });\n\n // Search with high topK to scan through many articles\n const results = await index.query({\n vector: dummyEmbedding.data[0].embedding,\n topK: 100,\n includeMetadata: true,\n filter: filter,\n });\n\n // Filter results by matching metrics\n const matchedResults = results.matches.filter(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n\n // Check if any key_metrics contain our search metrics\n return keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return metrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n });\n\n return matchedResults;\n}\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Analyze query for temporal indicators\n const dateRange = analyzeQueryTemporal(query);\n console.log(\"Date range analysis:\", dateRange);\n\n // Extract metrics from query\n const queryMetrics = extractMetricsFromQuery(query);\n console.log(\"Extracted metrics:\", queryMetrics);\n\n // If we have specific metrics, try metric-based search first\n if (queryMetrics.length > 0) {\n console.log(\"Searching by metrics first...\");\n const metricResults = await searchByMetrics(queryMetrics, dateRange);\n if (metricResults.length > 0) {\n console.log(`Found ${metricResults.length} results matching metrics`);\n // If we found good metric matches, prioritize them\n // But also do a regular search to complement\n }\n }\n\n // Adjust topK based on query type\n if (dateRange.isLatest) {\n topK = Math.max(topK, 50); // Get many results for latest queries\n } else if (dateRange.isDefault) {\n topK = Math.max(topK, 30); // Get more results for default recent queries\n } else {\n topK = Math.max(topK, 20); // Standard amount for specific periods\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n console.log(`Date range: ${dateRange.startDate} to ${dateRange.endDate}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone with date range\n let pineconeFilter = {\n publish_date_iso: {\n $gte: dateRange.startDate,\n $lte: dateRange.endDate,\n },\n };\n\n // Add other filters if provided\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n console.log(\"Pinecone filter:\", JSON.stringify(pineconeFilter));\n\n // Search Pinecone with date filter\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: pineconeFilter,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n\n // If we got metric results earlier, merge them with vector search results\n // Prioritize metric matches by boosting their scores\n if (queryMetrics.length > 0) {\n searchResults.matches.forEach(match => {\n const keyMetrics = JSON.parse(match.metadata.key_metrics || \"[]\");\n const hasMetricMatch = keyMetrics.some(km => {\n const metricStr = `${km.metric} ${km.context}`.toLowerCase();\n return queryMetrics.some(searchMetric => metricStr.includes(searchMetric));\n });\n\n if (hasMetricMatch) {\n // Boost score for metric matches\n match.score = match.score * 1.5;\n console.log(`Boosted score for article ${match.metadata.article_id} due to metric match`);\n }\n });\n }\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles\", {}, 50);\n const dates = results.map(r => ({\n date: r.metadata.publish_date,\n date_iso: r.metadata.publish_date_iso,\n parsed: parseMEESDate(r.metadata.publish_date).toISOString(),\n title: r.metadata.title ? r.metadata.title.substring(0, 60) : \"No title\",\n article_id: r.metadata.article_id,\n }));\n\n // Sort by ISO date\n dates.sort((a, b) => new Date(b.date_iso || b.parsed).getTime() - new Date(a.date_iso || a.parsed).getTime());\n\n return new Response(\n JSON.stringify({\n response: `Found ${dates.length} articles. Latest 10:\\n\\n`\n + dates.slice(0, 10).map((d, i) =>\n `${i + 1}. ${d.title}\\n Date: ${d.date} (ISO: ${d.date_iso || \"missing\"})\\n ID: ${d.article_id}`\n ).join(\"\\n\\n\"),\n debug_data: dates.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone with enhanced date-aware search\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Don't truncate text - the assistant needs full context\n let textContent = meta.text || \"\";\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso,\n text: textContent,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Sort by date using ISO date field - newest first\n console.log(\"Sorting results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = new Date(a.publish_date_iso || a.date);\n const dateB = new Date(b.publish_date_iso || b.date);\n return dateB.getTime() - dateA.getTime();\n });\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n date_iso: r.publish_date_iso,\n score: r.score,\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",12,[142],{"_143":144,"_99":261,"_104":354,"_146":132,"_161":355},["D",1748611256644],[356],{"_102":357,"_146":132,"_143":144,"_95":142,"_154":358,"_152":153,"_99":359,"_145":-5,"_104":354,"_147":261,"_157":-5,"_148":27,"_156":27,"_98":360},"eb89869a-3d58-11f0-9f32-76b3cceeab13","// MEES Assistant API Endpoint with v2 API - IMPROVED VERSION\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n};\n\n// Enhanced date parsing function with better error handling\nfunction parseMEESDate(dateStr) {\n if (!dateStr) return new Date(0);\n\n try {\n // Handle both \"5/23/25 20:00\" and ISO formats\n if (dateStr.includes(\"T\")) {\n return new Date(dateStr);\n }\n\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Enhanced search function with better chronological handling\nasync function searchPinecone(query, filters = {}, topK = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Determine if this is a time-sensitive query\n const lowerQuery = query.toLowerCase();\n const isTimeSensitive = lowerQuery.includes(\"latest\")\n || lowerQuery.includes(\"recent\")\n || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"current\")\n || lowerQuery.includes(\"now\")\n || lowerQuery.includes(\"2025\")\n || lowerQuery.includes(\"2024\");\n\n // For time-sensitive queries, get more results to ensure we capture recent articles\n if (isTimeSensitive) {\n topK = Math.max(topK, 50); // Increased from 40\n } else {\n topK = Math.max(topK, 25); // Increased from 20\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}, time-sensitive=${isTimeSensitive}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone\n let pineconeFilter = {};\n\n // Add date filtering for time-sensitive queries\n if (isTimeSensitive) {\n // Get articles from the last 6 months by default for \"latest\" queries\n const sixMonthsAgo = new Date();\n sixMonthsAgo.setMonth(sixMonthsAgo.getMonth() - 6);\n\n // Note: This requires your Pinecone index to have a numeric timestamp field\n // You might need to add this during indexing\n // pineconeFilter.publish_timestamp = { \"$gte\": sixMonthsAgo.getTime() };\n }\n\n // Apply other filters\n if (filters.companies && filters.companies.length > 0) {\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n // For time-sensitive queries, also try to boost the query with temporal terms\n let enhancedQuery = query;\n if (isTimeSensitive && !query.includes(\"2025\") && !query.includes(\"2024\")) {\n enhancedQuery = query + \" 2025 2024 recent latest\";\n }\n\n // Search Pinecone with enhanced query for time-sensitive searches\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n\n // If time-sensitive, also do a secondary search with explicit date terms\n if (isTimeSensitive && searchResults.matches.length < 10) {\n console.log(\"Performing secondary search with explicit temporal terms...\");\n\n const temporalQuery = query + \" 2025 latest recent current\";\n const temporalEmbedding = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: temporalQuery,\n });\n\n const temporalResults = await index.query({\n vector: temporalEmbedding.data[0].embedding,\n topK: 20,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n // Merge results, avoiding duplicates\n const existingIds = new Set(searchResults.matches.map(m => m.id));\n const newMatches = temporalResults.matches.filter(m => !existingIds.has(m.id));\n searchResults.matches.push(...newMatches);\n }\n\n return searchResults.matches;\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces)\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces)\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\n// Enhanced formatting function to ensure chronological context\nfunction formatResultsForAssistant(results) {\n // Group results by article to avoid duplicate information\n const articleMap = new Map();\n\n results.forEach(result => {\n const articleId = result.article_id;\n if (!articleMap.has(articleId)) {\n articleMap.set(articleId, {\n article_id: articleId,\n title: result.title,\n date: result.date,\n publish_date_iso: result.publish_date_iso,\n chunks: [],\n score: result.score,\n parsed_date: parseMEESDate(result.date),\n });\n }\n\n articleMap.get(articleId).chunks.push({\n chunk_num: result.chunk_num,\n text: result.text,\n score: result.score,\n media_in_text: result.media_in_text,\n key_metrics: result.key_metrics,\n section: result.section,\n });\n });\n\n // Convert to array and sort by date (newest first)\n const articles = Array.from(articleMap.values());\n articles.sort((a, b) => b.parsed_date.getTime() - a.parsed_date.getTime());\n\n // Add chronological context\n const now = new Date();\n articles.forEach(article => {\n const daysDiff = Math.floor((now - article.parsed_date) / (1000 * 60 * 60 * 24));\n\n if (daysDiff < 7) {\n article.recency = \"Published this week\";\n } else if (daysDiff < 30) {\n article.recency = \"Published this month\";\n } else if (daysDiff < 90) {\n article.recency = \"Published in the last 3 months\";\n } else if (daysDiff < 180) {\n article.recency = \"Published in the last 6 months\";\n } else {\n article.recency = `Published ${Math.floor(daysDiff / 30)} months ago`;\n }\n });\n\n return articles;\n}\n\nexport default async function(req) {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"energy production 2025\", {}, 50);\n const formattedResults = formatResultsForAssistant(results);\n\n return new Response(\n JSON.stringify({\n response: `Found ${formattedResults.length} unique articles. Latest 10:\\n\\n`\n + formattedResults.slice(0, 10).map((a, i) =>\n `${i + 1}. ${a.title}\\n Date: ${a.date} (${a.recency})\\n ID: ${a.article_id}`\n ).join(\"\\n\\n\"),\n debug_data: formattedResults.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results with chronological awareness\n const formattedResults = results.map((match) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso || meta.publish_date,\n text: meta.text || \"\",\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // ALWAYS sort by date - newest first\n console.log(\"Sorting all results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = parseMEESDate(a.date);\n const dateB = parseMEESDate(b.date);\n return dateB.getTime() - dateA.getTime(); // Newest first\n });\n\n // Use the enhanced formatting function\n const enhancedResults = formatResultsForAssistant(formattedResults);\n\n console.log(\n \"Enhanced results (top 5):\",\n enhancedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n recency: r.recency,\n })),\n );\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(enhancedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // 2 minutes\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000));\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders },\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders },\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",11,[142],{"_143":144,"_99":270,"_104":362,"_146":132,"_161":363},["D",1748610974403],[364],{"_102":365,"_146":132,"_143":144,"_95":142,"_154":366,"_152":153,"_99":367,"_145":-5,"_104":362,"_147":270,"_157":-5,"_148":27,"_156":27,"_98":368},"434fb33c-3d58-11f0-a502-76b3cceeab13","// Enhanced Pinecone Indexing Script for MEES Articles\n// This shows how to improve your indexing to support better temporal search\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\n// Enhanced date parsing with multiple format support\nfunction parseDate(dateStr) {\n if (!dateStr) return null;\n\n try {\n // Handle ISO format\n if (dateStr.includes(\"T\")) {\n return new Date(dateStr);\n }\n\n // Handle \"M/D/YY HH:MM\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return null;\n }\n}\n\n// Extract temporal context from text\nfunction extractTemporalContext(text, publishDate) {\n const context = {\n mentions_current_year: false,\n mentions_future: false,\n mentions_past_comparison: false,\n temporal_keywords: [],\n };\n\n const currentYear = new Date(publishDate).getFullYear();\n\n // Check for current year mentions\n const yearRegex = new RegExp(`\\\\b${currentYear}\\\\b`, \"g\");\n context.mentions_current_year = yearRegex.test(text);\n\n // Check for future projections\n const futurePatterns = /\\b(forecast|project|expect|will|planned|upcoming|next year|202[6-9])\\b/gi;\n context.mentions_future = futurePatterns.test(text);\n\n // Check for historical comparisons\n const pastPatterns = /\\b(compared to|versus|vs|from|since|last year|previously|historical)\\b/gi;\n context.mentions_past_comparison = pastPatterns.test(text);\n\n // Extract temporal keywords\n const temporalWords = text.match(/\\b(today|yesterday|recent|latest|current|now|this week|this month|this year)\\b/gi);\n if (temporalWords) {\n context.temporal_keywords = [...new Set(temporalWords.map(w => w.toLowerCase()))];\n }\n\n return context;\n}\n\n// Enhanced metric extraction with categorization\nfunction extractEnhancedMetrics(text) {\n const metrics = [];\n\n // Production metrics (bcf/d, mn t/y, b/d, etc.)\n const productionRegex = /(\\d+(?:\\.\\d+)?)\\s*(bcf\\/d|mn t\\/y|b\\/d|MW|GW|mcm\\/d)/gi;\n let match;\n while ((match = productionRegex.exec(text)) !== null) {\n metrics.push({\n value: match[1],\n unit: match[2],\n type: \"production\",\n context: text.substring(Math.max(0, match.index - 50), Math.min(text.length, match.index + 50)),\n });\n }\n\n // Price metrics\n const priceRegex = /\\$(\\d+(?:\\.\\d+)?)\\s*(\\/barrel|\\/mn BTU|\\/t|bn|mn)?/gi;\n while ((match = priceRegex.exec(text)) !== null) {\n metrics.push({\n value: match[1],\n unit: match[2] || \"\",\n type: \"price\",\n context: text.substring(Math.max(0, match.index - 50), Math.min(text.length, match.index + 50)),\n });\n }\n\n // Percentage changes\n const percentRegex = /(\\d+(?:\\.\\d+)?)\\s*%\\s*(increase|decrease|up|down|growth|decline)/gi;\n while ((match = percentRegex.exec(text)) !== null) {\n metrics.push({\n value: match[1],\n unit: \"%\",\n type: \"change\",\n direction: match[2],\n context: text.substring(Math.max(0, match.index - 50), Math.min(text.length, match.index + 50)),\n });\n }\n\n return metrics;\n}\n\n// Main indexing function with temporal enhancements\nasync function indexArticleWithTemporalData(article, chunkSize = 1000) {\n const index = pinecone.index(\"mees-articles\");\n\n const publishDate = parseDate(article.publish_date);\n if (!publishDate) {\n console.error(\"Invalid date for article:\", article.article_id);\n return;\n }\n\n // Add temporal metadata\n const baseMetadata = {\n article_id: article.article_id,\n title: article.title,\n publish_date: article.publish_date,\n publish_date_iso: publishDate.toISOString(),\n\n // New temporal fields\n publish_timestamp: publishDate.getTime(),\n publish_year: publishDate.getFullYear(),\n publish_month: publishDate.getMonth() + 1,\n publish_quarter: Math.ceil((publishDate.getMonth() + 1) / 3),\n publish_week: getWeekNumber(publishDate),\n days_ago: Math.floor((Date.now() - publishDate.getTime()) / (1000 * 60 * 60 * 24)),\n\n // Content analysis flags\n has_production_data: /\\b(production|output|capacity|bcf\\/d|b\\/d)\\b/i.test(article.text),\n has_price_data: /\\$\\d+|\\b(price|cost|value)\\b/i.test(article.text),\n has_import_export_data: /\\b(import|export|cargo|shipment|delivery)\\b/i.test(article.text),\n has_infrastructure_data: /\\b(pipeline|terminal|refinery|plant|facility)\\b/i.test(article.text),\n\n // Original metadata\n primary_country: article.primary_country,\n section: article.section,\n companies: JSON.stringify(article.companies || []),\n locations: JSON.stringify(article.locations || []),\n key_topics: JSON.stringify(article.key_topics || []),\n\n // Enhanced metrics with categorization\n metrics_summary: JSON.stringify({\n production: extractEnhancedMetrics(article.text).filter(m => m.type === \"production\"),\n price: extractEnhancedMetrics(article.text).filter(m => m.type === \"price\"),\n change: extractEnhancedMetrics(article.text).filter(m => m.type === \"change\"),\n }),\n };\n\n // Extract temporal context\n const temporalContext = extractTemporalContext(article.text, publishDate);\n baseMetadata.temporal_context = JSON.stringify(temporalContext);\n\n // Create chunks with overlap for better context\n const chunks = createOverlappingChunks(article.text, chunkSize, 200); // 200 char overlap\n\n const vectors = [];\n\n for (let i = 0; i < chunks.length; i++) {\n const chunk = chunks[i];\n\n // Create embedding with temporal context in mind\n // For recent articles mentioning \"latest\" or \"current\", boost those terms\n let embeddingText = chunk.text;\n if (baseMetadata.days_ago < 30 && temporalContext.temporal_keywords.length > 0) {\n embeddingText = `Latest update: ${chunk.text}`;\n }\n\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: embeddingText,\n });\n\n const chunkMetadata = {\n ...baseMetadata,\n chunk_num: i + 1,\n total_chunks: chunks.length,\n chunk_text: chunk.text,\n chunk_start_char: chunk.start,\n chunk_end_char: chunk.end,\n\n // Chunk-specific temporal indicators\n chunk_has_date_reference: /\\b(jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\\w*\\s+\\d{2,4}\\b/i.test(chunk.text),\n chunk_has_temporal_keyword: temporalContext.temporal_keywords.some(kw => chunk.text.toLowerCase().includes(kw)),\n };\n\n vectors.push({\n id: `${article.article_id}_chunk_${i + 1}`,\n values: embeddingResponse.data[0].embedding,\n metadata: chunkMetadata,\n });\n }\n\n // Upsert vectors in batches\n const batchSize = 100;\n for (let i = 0; i < vectors.length; i += batchSize) {\n const batch = vectors.slice(i, i + batchSize);\n await index.upsert(batch);\n console.log(`Upserted ${batch.length} vectors for article ${article.article_id}`);\n }\n}\n\n// Helper function to create overlapping chunks\nfunction createOverlappingChunks(text, chunkSize, overlap) {\n const chunks = [];\n let start = 0;\n\n while (start < text.length) {\n let end = start + chunkSize;\n\n // Try to break at sentence boundary\n if (end < text.length) {\n const nextPeriod = text.indexOf(\".\", end);\n if (nextPeriod !== -1 && nextPeriod < end + 100) {\n end = nextPeriod + 1;\n }\n }\n\n chunks.push({\n text: text.substring(start, Math.min(end, text.length)),\n start: start,\n end: Math.min(end, text.length),\n });\n\n start = end - overlap;\n if (start >= text.length - overlap) break;\n }\n\n return chunks;\n}\n\n// Helper function to get week number\nfunction getWeekNumber(date) {\n const firstDayOfYear = new Date(date.getFullYear(), 0, 1);\n const pastDaysOfYear = (date - firstDayOfYear) / 86400000;\n return Math.ceil((pastDaysOfYear + firstDayOfYear.getDay() + 1) / 7);\n}\n\n// Example: Create a daily indexing job that updates recency\nasync function updateRecencyMetadata() {\n const index = pinecone.index(\"mees-articles\");\n\n // This would need to be run as a scheduled job\n // It updates the days_ago field for all articles\n console.log(\"Updating recency metadata is a batch operation that requires careful planning\");\n\n // In practice, you might:\n // 1. Query all unique article IDs\n // 2. Calculate new days_ago values\n // 3. Update metadata in batches\n\n // For now, this is just a placeholder to show the concept\n}\n\n// Export for use in your indexing pipeline\nexport {\n extractEnhancedMetrics,\n extractTemporalContext,\n indexArticleWithTemporalData,\n parseDate,\n updateRecencyMetadata,\n};",10,[142],{"_143":144,"_99":279,"_104":370,"_146":132,"_161":371},["D",1748609773054],[372],{"_102":373,"_146":132,"_143":144,"_95":142,"_154":374,"_152":153,"_99":375,"_145":-5,"_104":370,"_147":279,"_157":-5,"_148":27,"_156":27,"_98":376},"773f6b36-3d55-11f0-9d1e-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Always get more results to ensure we catch recent articles\n // For \"latest\" queries, get even more\n const lowerQuery = query.toLowerCase();\n if (lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")) {\n topK = Math.max(topK, 40); // Get many results for time-sensitive queries\n } else {\n topK = Math.max(topK, 20); // Default minimum for chronological coverage\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone\n let pineconeFilter = {};\n\n if (filters.companies && filters.companies.length > 0) {\n // Note: This won't work with JSON strings, would need preprocessing\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles 2025\", {}, 30);\n const dates = results.map(r => ({\n date: r.metadata.publish_date,\n parsed: parseMEESDate(r.metadata.publish_date).toISOString(),\n title: r.metadata.title ? r.metadata.title.substring(0, 60) : \"No title\",\n article_id: r.metadata.article_id,\n }));\n\n // Sort by parsed date\n dates.sort((a, b) => new Date(b.parsed).getTime() - new Date(a.parsed).getTime());\n\n return new Response(\n JSON.stringify({\n response: `Found ${dates.length} articles. Latest 10:\\n\\n`\n + dates.slice(0, 10).map((d, i) =>\n `${i + 1}. ${d.title}\\n Date: ${d.date} (${\n new Date(d.parsed).toLocaleDateString()\n })\\n ID: ${d.article_id}`\n ).join(\"\\n\\n\"),\n debug_data: dates.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Don't truncate text - the assistant needs full context\n let textContent = meta.text || \"\";\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso || meta.publish_date,\n text: textContent,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // ALWAYS sort by date - newest first for chronological awareness\n console.log(\"Sorting all results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = parseMEESDate(a.date);\n const dateB = parseMEESDate(b.date);\n return dateB.getTime() - dateA.getTime(); // Newest first\n });\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n parsedDate: parseMEESDate(r.date).toISOString(),\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",9,[142],{"_143":144,"_99":287,"_104":378,"_146":132,"_161":379},["D",1748609028854],[380],{"_102":381,"_146":132,"_143":144,"_95":142,"_154":382,"_152":153,"_99":383,"_145":-5,"_104":378,"_147":287,"_157":-5,"_148":27,"_156":27,"_98":384},"bbab8036-3d53-11f0-aa1b-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Always get more results to ensure we catch recent articles\n // For \"latest\" queries, get even more\n const lowerQuery = query.toLowerCase();\n if (lowerQuery.includes(\"latest\") || lowerQuery.includes(\"most recent\") || lowerQuery.includes(\"newest\")) {\n topK = Math.max(topK, 40); // Get many results for time-sensitive queries\n } else {\n topK = Math.max(topK, 20); // Default minimum for chronological coverage\n }\n\n console.log(`Searching for: \"${query}\" with topK=${topK}`);\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone\n let pineconeFilter = {};\n\n if (filters.companies && filters.companies.length > 0) {\n // Note: This won't work with JSON strings, would need preprocessing\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles 2025\", {}, 30);\n const dates = results.map(r => ({\n date: r.metadata.publish_date,\n parsed: parseMEESDate(r.metadata.publish_date).toISOString(),\n title: r.metadata.title ? r.metadata.title.substring(0, 60) : \"No title\",\n article_id: r.metadata.article_id,\n }));\n\n // Sort by parsed date\n dates.sort((a, b) => new Date(b.parsed).getTime() - new Date(a.parsed).getTime());\n\n return new Response(\n JSON.stringify({\n response: `Found ${dates.length} articles. Latest 10:\\n\\n`\n + dates.slice(0, 10).map((d, i) =>\n `${i + 1}. ${d.title}\\n Date: ${d.date} (${\n new Date(d.parsed).toLocaleDateString()\n })\\n ID: ${d.article_id}`\n ).join(\"\\n\\n\"),\n debug_data: dates.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Don't truncate text - the assistant needs full context\n let textContent = meta.text || \"\";\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso || meta.publish_date,\n text: textContent,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // Don't truncate text - the assistant needs full context\n let textContent = meta.text || \"\";\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso || meta.publish_date,\n text: textContent,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n section: meta.section,\n country: meta.primary_country,\n };\n });\n\n // ALWAYS sort by date - newest first for chronological awareness\n console.log(\"Sorting all results by date (newest first)...\");\n formattedResults.sort((a, b) => {\n const dateA = parseMEESDate(a.date);\n const dateB = parseMEESDate(b.date);\n return dateB.getTime() - dateA.getTime(); // Newest first\n });\n\n console.log(\n \"Results after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n parsedDate: parseMEESDate(r.date).toISOString(),\n })),\n );\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",8,[142],{"_143":144,"_99":295,"_104":386,"_146":132,"_161":387},["D",1748607538675],[388],{"_102":389,"_146":132,"_143":144,"_95":142,"_154":390,"_152":153,"_99":391,"_145":-5,"_104":386,"_147":295,"_157":-5,"_148":27,"_156":27,"_98":392},"43742a12-3d50-11f0-ab46-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Enhance query for time-based searches\n let enhancedQuery = query;\n const lowerQuery = query.toLowerCase();\n const isTimeBasedQuery = lowerQuery.includes(\"latest\")\n || lowerQuery.includes(\"recent\")\n || lowerQuery.includes(\"newest\")\n || lowerQuery.includes(\"published\");\n\n const isDateSpecificQuery = lowerQuery.includes(\"may 2025\")\n || lowerQuery.includes(\"april 2025\")\n || lowerQuery.includes(\"march 2025\")\n || lowerQuery.includes(\"2025\");\n\n if (isTimeBasedQuery || isDateSpecificQuery) {\n // For time-based queries, bias toward 2025\n if (!query.includes(\"2025\")) {\n enhancedQuery = query + \" 2025 energy MEES\";\n }\n console.log(\"Enhanced query for recency:\", enhancedQuery);\n // Increase topK to get more results to sort through\n topK = Math.min(Math.max(topK, 20), 50); // Cap at 50 to avoid too much data\n }\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: enhancedQuery,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone\n let pineconeFilter = {};\n\n if (filters.companies && filters.companies.length > 0) {\n // Note: This won't work with JSON strings, would need preprocessing\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n console.log(`Search returned ${searchResults.matches.length} results`);\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Debug endpoint for testing date queries\n if (message === \"DEBUG_DATES\") {\n const results = await searchPinecone(\"MEES articles 2025\", {}, 30);\n const dates = results.map(r => ({\n date: r.metadata.publish_date,\n parsed: parseMEESDate(r.metadata.publish_date).toISOString(),\n title: r.metadata.title ? r.metadata.title.substring(0, 60) : \"No title\",\n article_id: r.metadata.article_id,\n }));\n\n // Sort by parsed date\n dates.sort((a, b) => new Date(b.parsed).getTime() - new Date(a.parsed).getTime());\n\n return new Response(\n JSON.stringify({\n response: `Found ${dates.length} articles. Latest 10:\\n\\n`\n + dates.slice(0, 10).map((d, i) =>\n `${i + 1}. ${d.title}\\n Date: ${d.date} (${\n new Date(d.parsed).toLocaleDateString()\n })\\n ID: ${d.article_id}`\n ).join(\"\\n\\n\"),\n debug_data: dates.slice(0, 10),\n threadId: \"debug\",\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n console.log(`Handling ${toolCalls.length} tool calls`);\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n console.log(`Tool call: ${name}`);\n\n const parsedArgs = JSON.parse(args);\n console.log(`Arguments:`, JSON.stringify(parsedArgs).substring(0, 200));\n\n if (name === \"search_articles\") {\n try {\n console.log(\"Searching Pinecone...\");\n // Search Pinecone\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n console.log(`Found ${results.length} results from Pinecone`);\n\n // Check if this is a time-based query\n const isTimeBasedQuery = parsedArgs.query.toLowerCase().includes(\"latest\")\n || parsedArgs.query.toLowerCase().includes(\"recent\")\n || parsedArgs.query.toLowerCase().includes(\"newest\")\n || parsedArgs.query.toLowerCase().includes(\"published\");\n\n // Format results for the assistant\n let formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n // For time-based queries, truncate text to reduce processing time\n let textContent = meta.text || \"\";\n if (isTimeBasedQuery && textContent.length > 500) {\n textContent = textContent.substring(0, 500) + \"... [truncated for overview]\";\n }\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso || meta.publish_date,\n text: textContent,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n // Add summary fields for time-based queries\n summary: isTimeBasedQuery\n ? {\n hasCharts: meta.has_charts || false,\n hasTables: meta.has_tables || false,\n hasMaps: meta.has_maps || false,\n section: meta.section,\n country: meta.primary_country,\n }\n : null,\n };\n });\n\n // Sort by date if the query seems time-based\n if (isTimeBasedQuery) {\n console.log(\"Sorting results by date...\");\n formattedResults.sort((a, b) => {\n // Parse dates using our custom function\n const dateA = parseMEESDate(a.date);\n const dateB = parseMEESDate(b.date);\n return dateB.getTime() - dateA.getTime(); // Newest first\n });\n\n console.log(\n \"Top 5 articles after sorting:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n parsedDate: parseMEESDate(r.date).toISOString(),\n })),\n );\n\n // If we got extra results for sorting, trim back to requested topK\n if (formattedResults.length > (parsedArgs.top_k || 5)) {\n formattedResults = formattedResults.slice(0, parsedArgs.top_k || 5);\n }\n }\n\n console.log(`Returning ${formattedResults.length} formatted results`);\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } catch (searchError) {\n console.error(\"Error in search_articles:\", searchError);\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({\n error: \"Search failed\",\n message: searchError.message,\n }),\n });\n }\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 120000; // Increased to 120 seconds (2 minutes)\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",7,[142],{"_143":144,"_99":303,"_104":394,"_146":132,"_161":395},["D",1748606769182],[396],{"_102":397,"_146":132,"_143":144,"_95":142,"_154":398,"_152":153,"_99":399,"_145":-5,"_104":394,"_147":303,"_157":-5,"_148":27,"_156":27,"_98":400},"78ccda08-3d4e-11f0-834e-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Enhance query for time-based searches\n let enhancedQuery = query;\n const isTimeBasedQuery = query.toLowerCase().includes(\"latest\")\n || query.toLowerCase().includes(\"recent\")\n || query.toLowerCase().includes(\"newest\");\n\n if (isTimeBasedQuery) {\n // Add 2025 to the query to bias toward recent articles\n enhancedQuery = query + \" 2025\";\n console.log(\"Enhanced query for recency:\", enhancedQuery);\n // Increase topK to get more results to sort through\n topK = Math.max(topK, 20);\n }\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: enhancedQuery,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone\n let pineconeFilter = {};\n\n if (filters.companies && filters.companies.length > 0) {\n // Note: This won't work with JSON strings, would need preprocessing\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n return searchResults.matches;\n}\n\n// Function to parse MEES date format \"5/23/25 20:00\" to proper Date object\nfunction parseMEESDate(dateStr: string): Date {\n if (!dateStr) return new Date(0); // Return epoch for missing dates\n\n try {\n // Parse \"5/23/25 20:00\" format\n const [datePart, timePart] = dateStr.split(\" \");\n const [month, day, year] = datePart.split(\"/\");\n const [hour, minute] = timePart ? timePart.split(\":\") : [\"0\", \"0\"];\n\n // Convert 2-digit year to 4-digit (assuming 20xx)\n const fullYear = parseInt(year) < 100 ? 2000 + parseInt(year) : parseInt(year);\n\n return new Date(fullYear, parseInt(month) - 1, parseInt(day), parseInt(hour), parseInt(minute));\n } catch (e) {\n console.error(\"Error parsing date:\", dateStr, e);\n return new Date(0);\n }\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n const parsedArgs = JSON.parse(args);\n\n if (name === \"search_articles\") {\n // Search Pinecone\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n // Format results for the assistant\n const formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n publish_date_iso: meta.publish_date_iso || meta.publish_date, // Include ISO date if available\n text: meta.text,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n };\n });\n\n // Sort by date if the query seems time-based\n if (\n parsedArgs.query.toLowerCase().includes(\"latest\")\n || parsedArgs.query.toLowerCase().includes(\"recent\")\n || parsedArgs.query.toLowerCase().includes(\"newest\")\n ) {\n formattedResults.sort((a, b) => {\n // Parse dates using our custom function\n const dateA = parseMEESDate(a.date);\n const dateB = parseMEESDate(b.date);\n return dateB.getTime() - dateA.getTime(); // Newest first\n });\n\n console.log(\"Sorted results by date - newest first\");\n console.log(\n \"Top 5 articles:\",\n formattedResults.slice(0, 5).map(r => ({\n title: r.title.substring(0, 50) + \"...\",\n date: r.date,\n parsedDate: parseMEESDate(r.date).toISOString(),\n })),\n );\n\n // If we got extra results for sorting, trim back to requested topK\n if (formattedResults.length > (parsedArgs.top_k || 5)) {\n formattedResults = formattedResults.slice(0, parsedArgs.top_k || 5);\n }\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 60000; // 60 seconds\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",6,[142],{"_143":144,"_99":311,"_104":402,"_146":132,"_161":403},["D",1748606079289],[404],{"_102":405,"_146":132,"_143":144,"_95":142,"_154":406,"_152":153,"_99":407,"_145":-5,"_104":402,"_147":311,"_157":-5,"_148":27,"_156":27,"_98":408},"dd976ebe-3d4c-11f0-809b-76b3cceeab13","// MEES Assistant API Endpoint with v2 API\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\n// Use OpenAI client only for embeddings\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n});\n\n// Initialize Pinecone\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone\n let pineconeFilter = {};\n\n if (filters.companies && filters.companies.length > 0) {\n // Note: This won't work with JSON strings, would need preprocessing\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n return searchResults.matches;\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n if (!OPENAI_API_KEY) {\n throw new Error(\"OPENAI_API_KEY not found in environment variables\");\n }\n\n if (!ASSISTANT_ID) {\n throw new Error(\"ASSISTANT_ID not found in environment variables\");\n }\n\n // IMPORTANT: All OpenAI API calls use these headers with assistants=v2\n const baseHeaders = {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\", // This is the critical header for v2 API\n };\n\n console.log(\"Assistant ID:\", ASSISTANT_ID);\n console.log(\"Using OpenAI-Beta: assistants=v2\");\n\n // Create or retrieve thread using raw fetch with v2 headers\n let thread;\n if (threadId) {\n const threadResponse = await fetch(`https://api.openai.com/v1/threads/${threadId}`, {\n headers: baseHeaders,\n });\n if (!threadResponse.ok) {\n throw new Error(`Thread retrieval failed: ${await threadResponse.text()}`);\n }\n thread = await threadResponse.json();\n } else {\n const threadResponse = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({}),\n });\n if (!threadResponse.ok) {\n const error = await threadResponse.text();\n throw new Error(`Thread creation failed: ${error}`);\n }\n thread = await threadResponse.json();\n }\n\n // Add message to thread with v2 headers\n const messageResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/messages`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n role: \"user\",\n content: message,\n }),\n });\n\n if (!messageResponse.ok) {\n throw new Error(`Failed to add message: ${await messageResponse.text()}`);\n }\n\n // Run the assistant with v2 headers\n const runResponse = await fetch(`https://api.openai.com/v1/threads/${thread.id}/runs`, {\n method: \"POST\",\n headers: baseHeaders,\n body: JSON.stringify({\n assistant_id: ASSISTANT_ID,\n }),\n });\n\n if (!runResponse.ok) {\n throw new Error(`Failed to run assistant: ${await runResponse.text()}`);\n }\n\n const run = await runResponse.json();\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n const parsedArgs = JSON.parse(args);\n\n if (name === \"search_articles\") {\n // Search Pinecone\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n // Format results for the assistant\n const formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n text: meta.text,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n };\n });\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion using raw fetch with v2 headers\n let runStatus = run;\n let allMediaDisplays = [];\n const maxWaitTime = 60000; // 60 seconds\n const startTime = Date.now();\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n // Check if we've exceeded max wait time\n if (Date.now() - startTime > maxWaitTime) {\n throw new Error(\"Assistant took too long to respond\");\n }\n\n await new Promise(resolve => setTimeout(resolve, 2000)); // Wait 2 seconds between checks\n\n const statusResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!statusResponse.ok) {\n throw new Error(`Failed to get run status: ${await statusResponse.text()}`);\n }\n\n runStatus = await statusResponse.json();\n console.log(\"Run status:\", runStatus.status);\n\n if (runStatus.status === \"requires_action\") {\n console.log(\"Assistant requires action - handling tool calls\");\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n console.log(`Number of tool calls: ${toolCalls.length}`);\n\n try {\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n // Submit tool outputs with v2 headers\n const toolOutputResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/runs/${run.id}/submit_tool_outputs`,\n {\n method: \"POST\",\n headers: baseHeaders, // v2 headers\n body: JSON.stringify({ tool_outputs: toolOutputs }),\n },\n );\n\n if (!toolOutputResponse.ok) {\n throw new Error(`Failed to submit tool outputs: ${await toolOutputResponse.text()}`);\n }\n } catch (toolError) {\n console.error(\"Error handling tool calls:\", toolError);\n // Continue anyway - the assistant might still provide a response\n }\n }\n }\n\n if (runStatus.status === \"failed\") {\n console.error(\"Run failed:\", runStatus.last_error);\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message || \"Unknown error\"}`);\n }\n\n // Get the assistant's response with v2 headers\n const messagesResponse = await fetch(\n `https://api.openai.com/v1/threads/${thread.id}/messages?order=desc&limit=20`,\n { headers: baseHeaders }, // v2 headers\n );\n\n if (!messagesResponse.ok) {\n throw new Error(`Failed to get messages: ${await messagesResponse.text()}`);\n }\n\n const messages = await messagesResponse.json();\n\n console.log(\"Total messages:\", messages.data.length);\n console.log(\"First 3 messages roles:\", messages.data.slice(0, 3).map(m => m.role));\n\n // Find the most recent assistant message\n let assistantMessage = null;\n for (const msg of messages.data) {\n console.log(`Message role: ${msg.role}, created_at: ${msg.created_at}`);\n if (msg.role === \"assistant\") {\n assistantMessage = msg;\n break;\n }\n }\n\n let response = \"\";\n if (assistantMessage && assistantMessage.content && assistantMessage.content.length > 0) {\n console.log(\"Assistant message content:\", JSON.stringify(assistantMessage.content));\n const content = assistantMessage.content[0];\n if (content && content.type === \"text\" && content.text && content.text.value) {\n response = String(content.text.value);\n }\n }\n\n // Ensure response is a string\n if (!response || typeof response !== \"string\") {\n console.log(\"No valid response found, using default\");\n response = \"I'm having trouble generating a response. Please try again.\";\n }\n\n console.log(\"Final response type:\", typeof response);\n console.log(\"Final response length:\", response.length);\n\n return new Response(\n JSON.stringify({\n response: String(response || \"No response from assistant\"),\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error details:\", error);\n console.error(\"Error stack:\", error.stack);\n return new Response(\n JSON.stringify({\n error: error.message || \"Unknown error occurred\",\n details: error.toString(),\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",5,[142],{"_143":144,"_99":319,"_104":410,"_146":132,"_161":411},["D",1748604948480],[412],{"_102":413,"_146":132,"_143":144,"_95":142,"_154":414,"_152":153,"_99":170,"_145":-5,"_104":410,"_147":319,"_157":-5,"_148":27,"_156":27,"_98":415},"3b93c344-3d4a-11f0-aa21-76b3cceeab13","export default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS, GET\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n // Test creating a thread with v2 API\n const response = await fetch(\"https://api.openai.com/v1/threads\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${Deno.env.get(\"OPENAI_API_KEY\")}`,\n \"Content-Type\": \"application/json\",\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n body: JSON.stringify({}),\n });\n\n const responseText = await response.text();\n\n // Try parsing as JSON\n let result;\n try {\n result = JSON.parse(responseText);\n } catch {\n result = { raw_text: responseText };\n }\n\n return new Response(\n JSON.stringify(\n {\n test: \"v2 API Test\",\n status: response.status,\n headers_sent: {\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n response: result,\n success: response.status === 200,\n },\n null,\n 2,\n ),\n {\n headers: { ...corsHeaders, \"Content-Type\": \"application/json\" },\n },\n );\n } catch (error) {\n return new Response(\n JSON.stringify({\n error: error.message,\n stack: error.stack,\n }),\n {\n status: 500,\n headers: { ...corsHeaders, \"Content-Type\": \"application/json\" },\n },\n );\n }\n}",[142],{"_143":144,"_99":327,"_104":417,"_146":132,"_161":418},["D",1748604389018],[419],{"_102":420,"_146":132,"_143":144,"_95":142,"_154":421,"_152":153,"_99":422,"_145":-5,"_104":417,"_147":327,"_157":-5,"_148":27,"_156":27,"_98":423},"ee1c9bd2-3d48-11f0-8002-76b3cceeab13","// MEES Assistant API Endpoint\n// Remember to add your environment variables in Val settings:\n// - OPENAI_API_KEY\n// - PINECONE_API_KEY\n// - ASSISTANT_ID\n\nimport { OpenAI } from \"https://deno.land/x/openai@v4.20.1/mod.ts\";\nimport { Pinecone } from \"https://esm.sh/@pinecone-database/pinecone@2.0.0\";\n\nconst openai = new OpenAI({\n apiKey: Deno.env.get(\"OPENAI_API_KEY\"),\n defaultHeaders: {\n \"OpenAI-Beta\": \"assistants=v2\",\n },\n});\n// Initialize Pinecone - newer version doesn't need environment\nconst pinecone = new Pinecone({\n apiKey: Deno.env.get(\"PINECONE_API_KEY\"),\n});\n\nconst ASSISTANT_ID = Deno.env.get(\"ASSISTANT_ID\");\n\n// Media ID to Title mapping - expand this with your actual mappings\nconst MEDIA_TITLES = {\n // Charts\n \"7489\": \"Egypt Gas Output & Consumption Balance\",\n \"7484\": \"The Seasonality Of Turkey's LNG Imports\",\n \"7485\": \"Egypt Gas Availability Will Increase Significantly This Summer\",\n \"7486\": \"Egypt LNG Imports By Origin\",\n \"7487\": \"Egypt Imports Of Israeli Gas Rise To Record\",\n \"7505\": \"Egypt Gas Deficit Widens Sharply\",\n \"7506\": \"Egypt Fuel Oil Burn For Power Rises\",\n // Add more mappings as needed\n // Tables\n \"5311\": \"Egypt Gas Infrastructure Summary\",\n // Maps\n \"1557\": \"Eastern Mediterranean Gas Fields\",\n \"1558\": \"Egypt LNG Terminal Locations\",\n};\n\nasync function searchPinecone(query: string, filters: any = {}, topK: number = 5) {\n const index = pinecone.index(\"mees-articles\");\n\n // Get embedding for query\n const embeddingResponse = await openai.embeddings.create({\n model: \"text-embedding-3-small\",\n input: query,\n });\n\n const queryEmbedding = embeddingResponse.data[0].embedding;\n\n // Build filter object for Pinecone\n let pineconeFilter = {};\n\n if (filters.companies && filters.companies.length > 0) {\n // Note: This won't work with JSON strings, would need preprocessing\n pineconeFilter.companies = { $in: filters.companies };\n }\n\n if (filters.has_charts === true) {\n pineconeFilter.has_charts = true;\n }\n\n if (filters.has_tables === true) {\n pineconeFilter.has_tables = true;\n }\n\n if (filters.has_maps === true) {\n pineconeFilter.has_maps = true;\n }\n\n if (filters.primary_country) {\n pineconeFilter.primary_country = filters.primary_country;\n }\n\n // Search Pinecone\n const searchResults = await index.query({\n vector: queryEmbedding,\n topK: topK,\n includeMetadata: true,\n filter: Object.keys(pineconeFilter).length > 0 ? pineconeFilter : undefined,\n });\n\n return searchResults.matches;\n}\n\n// Extract media from text using regex patterns\nfunction extractMediaFromText(text: string) {\n const media = [];\n\n // Extract charts (6 braces)\n const chartRegex = /\\{\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}\\}/g;\n let match;\n while ((match = chartRegex.exec(text)) !== null) {\n media.push({ id: match[1], type: \"chart\" });\n }\n\n // Extract maps (5 braces) - but not if already found as chart\n const mapRegex = /\\{\\{\\{\\{\\{(\\d+)\\}\\}\\}\\}\\}/g;\n while ((match = mapRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"map\" });\n }\n }\n\n // Extract tables (2 braces) - but not if already found\n const tableRegex = /\\{\\{(\\d+)\\}\\}/g;\n while ((match = tableRegex.exec(text)) !== null) {\n if (!media.find(m => m.id === match[1])) {\n media.push({ id: match[1], type: \"table\" });\n }\n }\n\n return media;\n}\n\nexport default async function(req: Request): Promise {\n const corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type\",\n };\n\n if (req.method === \"OPTIONS\") {\n return new Response(null, { headers: corsHeaders });\n }\n\n try {\n const { message, threadId } = await req.json();\n\n // Create or retrieve thread\n let thread;\n if (threadId) {\n thread = await openai.beta.threads.retrieve(threadId);\n } else {\n thread = await openai.beta.threads.create();\n }\n\n // Add message to thread\n await openai.beta.threads.messages.create(thread.id, {\n role: \"user\",\n content: message,\n });\n\n // Run the assistant\n const run = await openai.beta.threads.runs.create(thread.id, {\n assistant_id: ASSISTANT_ID,\n });\n\n // Function to handle tool calls\n const handleToolCalls = async (toolCalls: any[]) => {\n const toolOutputs = [];\n const mediaDisplays = [];\n\n for (const toolCall of toolCalls) {\n const { name, arguments: args } = toolCall.function;\n const parsedArgs = JSON.parse(args);\n\n if (name === \"search_articles\") {\n // Search Pinecone\n const results = await searchPinecone(\n parsedArgs.query,\n parsedArgs.filters || {},\n parsedArgs.top_k || 5,\n );\n\n // Format results for the assistant\n const formattedResults = results.map((match: any) => {\n const meta = match.metadata;\n\n // Extract media from text\n const mediaInText = extractMediaFromText(meta.text || \"\");\n\n return {\n article_id: meta.article_id,\n title: meta.title,\n date: meta.publish_date,\n text: meta.text,\n score: match.score,\n // Parse JSON strings back to arrays\n chunk_charts: JSON.parse(meta.chunk_charts || \"[]\"),\n chunk_tables: JSON.parse(meta.chunk_tables || \"[]\"),\n chunk_maps: JSON.parse(meta.chunk_maps || \"[]\"),\n all_charts: JSON.parse(meta.all_charts || \"[]\"),\n all_tables: JSON.parse(meta.all_tables || \"[]\"),\n all_maps: JSON.parse(meta.all_maps || \"[]\"),\n companies: JSON.parse(meta.companies || \"[]\"),\n key_metrics: JSON.parse(meta.key_metrics || \"[]\"),\n energy_assets: JSON.parse(meta.energy_assets || \"[]\"),\n chunk_num: meta.chunk_num,\n total_chunks: meta.total_chunks,\n media_in_text: mediaInText,\n };\n });\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify(formattedResults),\n });\n } else if (name === \"display_media\") {\n // Record media display requests\n const mediaItems = parsedArgs.media_items || [];\n\n for (const item of mediaItems) {\n mediaDisplays.push({\n id: item.id,\n type: item.type,\n title: item.title || MEDIA_TITLES[item.id] || `${item.type} ${item.id}`,\n article_id: item.article_id,\n relevance: item.relevance || \"\",\n });\n }\n\n toolOutputs.push({\n tool_call_id: toolCall.id,\n output: JSON.stringify({ status: \"media queued for display\", count: mediaItems.length }),\n });\n }\n }\n\n return { toolOutputs, mediaDisplays };\n };\n\n // Poll for completion\n let runStatus = await openai.beta.threads.runs.retrieve(thread.id, run.id);\n let allMediaDisplays = [];\n\n while (runStatus.status !== \"completed\" && runStatus.status !== \"failed\") {\n await new Promise(resolve => setTimeout(resolve, 1000));\n runStatus = await openai.beta.threads.runs.retrieve(thread.id, run.id);\n\n if (runStatus.status === \"requires_action\") {\n const toolCalls = runStatus.required_action.submit_tool_outputs.tool_calls;\n const { toolOutputs, mediaDisplays } = await handleToolCalls(toolCalls);\n allMediaDisplays = [...allMediaDisplays, ...mediaDisplays];\n\n await openai.beta.threads.runs.submitToolOutputs(thread.id, run.id, {\n tool_outputs: toolOutputs,\n });\n }\n }\n\n if (runStatus.status === \"failed\") {\n throw new Error(`Assistant run failed: ${runStatus.last_error?.message}`);\n }\n\n // Get the assistant's response\n const messages = await openai.beta.threads.messages.list(thread.id);\n const lastMessage = messages.data[0];\n const response = lastMessage.content[0].type === \"text\"\n ? lastMessage.content[0].text.value\n : \"\";\n\n return new Response(\n JSON.stringify({\n response,\n threadId: thread.id,\n media: allMediaDisplays,\n status: \"success\",\n }),\n {\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n } catch (error) {\n console.error(\"Error:\", error);\n return new Response(\n JSON.stringify({\n error: error.message,\n status: \"error\",\n }),\n {\n status: 500,\n headers: {\n ...corsHeaders,\n \"Content-Type\": \"application/json\",\n },\n },\n );\n }\n}",3,[142],{"_143":144,"_99":335,"_104":425,"_146":132,"_161":426},["D",1748604373555],[427],{"_102":428,"_146":132,"_143":144,"_95":142,"_154":429,"_152":153,"_99":430,"_145":-5,"_104":425,"_147":335,"_157":-5,"_148":27,"_156":27,"_98":431},"e4e518a0-3d48-11f0-ae26-76b3cceeab13","export default async function (req: Request): Promise {\n return Response.json({ ok: true })\n}",2,[142],{"_143":144,"_99":343,"_104":433,"_146":132,"_161":434},["D",1748604373331],[435],{"_102":436,"_146":132,"_143":144,"_95":142,"_154":68,"_152":153,"_99":167,"_145":-5,"_104":433,"_147":343,"_157":-5,"_148":27,"_156":27,"_98":437},"e4c2d1dc-3d48-11f0-b8f2-76b3cceeab13",[142],{"_143":144,"_99":351,"_104":439,"_146":132,"_161":440},["D",1748604370988],[441],{"_102":442,"_146":132,"_143":144,"_95":142,"_154":68,"_152":443,"_99":31,"_145":-5,"_104":439,"_147":351,"_157":-5,"_148":27,"_156":27,"_98":444},"e35cf782-3d48-11f0-8002-76b3cceeab13","script",[142],"page","hasNext","routes/_app","actionData","errors"]

History