Hi

\"\n});\n```\n\n## Val Town Utility Functions\n\nVal Town provides several utility functions to help with common project tasks.\n\n### Importing Utilities\n\nAlways import utilities with version pins to avoid breaking changes:\n\n```ts\nimport { parseProject, readFile, serveFile } from \"https://esm.town/v/std/utils@85-main/index.ts\";\n```\n\n### Available Utilities\n\n\n#### **serveFile** - Serve project files with proper content types\n\nFor example, in Hono:\n\n```ts\n// serve all files in frontend/ and shared/\napp.get(\"/frontend/*\", c => serveFile(c.req.path, import.meta.url));\napp.get(\"/shared/*\", c => serveFile(c.req.path, import.meta.url));\n```\n\n#### **readFile** - Read files from within the project:\n\n```ts\n// Read a file from the project\nconst fileContent = await readFile(\"/frontend/index.html\", import.meta.url);\n```\n\n#### **listFiles** - List all files in the project\n\n```ts\nconst files = await listFiles(import.meta.url);\n```\n\n#### **parseProject** - Extract information about the current project from import.meta.url\n\nThis is useful for including info for linking back to a val, ie in \"view source\" urls:\n\n```ts\nconst projectVal = parseProject(import.meta.url);\nconsole.log(projectVal.username); // Owner of the project\nconsole.log(projectVal.name); // Project name\nconsole.log(projectVal.version); // Version number\nconsole.log(projectVal.branch); // Branch name\nconsole.log(projectVal.links.self.project); // URL to the project page\n```\n\nHowever, it's *extremely importing* to note that `parseProject` and other Standard Library utilities ONLY RUN ON THE SERVER.\nIf you need access to this data on the client, run it in the server and pass it to the client by splicing it into the HTML page\nor by making an API request for it.\n\n## Val Town Platform Specifics\n\n- **Redirects:** Use `return new Response(null, { status: 302, headers: { Location: \"/place/to/redirect\" }})` instead of `Response.redirect` which is broken\n- **Images:** Avoid external images or base64 images. Use emojis, unicode symbols, or icon fonts/libraries instead\n- **AI Image:** To inline generate an AI image use: `

`\n- **Storage:** DO NOT use the Deno KV module for storage\n- **Browser APIs:** DO NOT use the `alert()`, `prompt()`, or `confirm()` methods\n- **Weather Data:** Use open-meteo for weather data (doesn't require API keys) unless otherwise specified\n- **View Source:** Add a view source link by importing & using `import.meta.url.replace(\"ems.sh\", \"val.town)\"` (or passing this data to the client) and include `target=\"_top\"` attribute\n- **Error Debugging:** Add `` to HTML to capture client-side errors\n- **Error Handling:** Only use try...catch when there's a clear local resolution; Avoid catches that merely log or return 500s. Let errors bubble up with full context\n- **Environment Variables:** Use `Deno.env.get('keyname')` when you need to, but generally prefer APIs that don't require keys\n- **Imports:** Use `https://esm.sh` for npm and Deno dependencies to ensure compatibility on server and browser\n- **Storage Strategy:** Only use backend storage if explicitly required; prefer simple static client-side sites\n- **React Configuration:** When using React libraries, pin versions with `?deps=react@18.2.0,react-dom@18.2.0` and start the file with `/** @jsxImportSource https://esm.sh/react@18.2.0 */`\n- Ensure all React dependencies and sub-dependencies are pinned to the same version\n- **Styling:** Default to using TailwindCSS via `` unless otherwise specified\n\n## Project Structure and Design Patterns\n\n### Recommended Directory Structure\n```\n├── backend/\n│ ├── database/\n│ │ ├── migrations.ts # Schema definitions\n│ │ ├── queries.ts # DB query functions\n│ │ └── README.md\n│ └── routes/ # Route modules\n│ ├── [route].ts\n│ └── static.ts # Static file serving\n│ ├── index.ts # Main entry point\n│ └── README.md\n├── frontend/\n│ ├── components/\n│ │ ├── App.tsx\n│ │ └── [Component].tsx\n│ ├── favicon.svg\n│ ├── index.html # Main HTML template\n│ ├── index.tsx # Frontend JS entry point\n│ ├── README.md\n│ └── style.css\n├── README.md\n└── shared/\n ├── README.md\n └── utils.ts # Shared types and functions\n```\n\n### Backend (Hono) Best Practices\n\n- Hono is the recommended API framework\n- Main entry point should be `backend/index.ts`\n- **Static asset serving:** Use the utility functions to read and serve project files:\n ```ts\n import { readFile, serveFile } from \"https://esm.town/v/std/utils@85-main/index.ts\";\n \n // serve all files in frontend/ and shared/\n app.get(\"/frontend/*\", c => serveFile(c.req.path, import.meta.url));\n app.get(\"/shared/*\", c => serveFile(c.req.path, import.meta.url));\n \n // For index.html, often you'll want to bootstrap with initial data\n app.get(\"/\", async c => {\n let html = await readFile(\"/frontend/index.html\", import.meta.url);\n \n // Inject data to avoid extra round-trips\n const initialData = await fetchInitialData();\n const dataScript = ``;\n \n html = html.replace(\"\", `${dataScript}`);\n return c.html(html);\n });\n ```\n- Create RESTful API routes for CRUD operations\n- Always include this snippet at the top-level Hono app to re-throwing errors to see full stack traces:\n ```ts\n // Unwrap Hono errors to see original error details\n app.onError((err, c) => {\n throw err;\n });\n ```\n\n### Database Patterns\n- Run migrations on startup or comment out for performance\n- Change table names when modifying schemas rather than altering\n- Export clear query functions with proper TypeScript typing\n\n## Common Gotchas and Solutions\n\n1. **Environment Limitations:** \n - Val Town runs on Deno in a serverless context, not Node.js\n - Code in `shared/` must work in both frontend and backend environments\n - Cannot use `Deno` keyword in shared code\n - Use `https://esm.sh` for imports that work in both environments\n\n2. **SQLite Peculiarities:**\n - Limited support for ALTER TABLE operations\n - Create new tables with updated schemas and copy data when needed\n - Always run table creation before querying\n\n3. **React Configuration:**\n - All React dependencies must be pinned to 18.2.0\n - Always include `@jsxImportSource https://esm.sh/react@18.2.0` at the top of React files\n - Rendering issues often come from mismatched React versions\n\n4. **File Handling:**\n - Val Town only supports text files, not binary\n - Use the provided utilities to read files across branches and forks\n - For files in the project, use `readFile` helpers\n\n5. **API Design:**\n - `fetch` handler is the entry point for HTTP vals\n - Run the Hono app with `export default app.fetch // This is the entry point for HTTP vals`\n\n",[196],"65892a86-b8e7-11f0-ba05-0224a6c84d84",".vtignore","65892b4e-b8e7-11f0-ba05-0224a6c84d84",9,["D",1762196442809],".git\n.vscode\n.cursorrules\n.DS_Store\nnode_modules\nvendor",[203],"7ecff464-b9ca-11f0-8b68-0224a6c84d84","utils","7ecff54a-b9ca-11f0-8b68-0224a6c84d84",49,["D",1762293981048],[210],"30dc0596-b901-11f0-a3b9-0224a6c84d84","testing","30dc0668-b901-11f0-a3b9-0224a6c84d84",28,["D",1762207521420],[216],"30c2711c-b901-11f0-b5de-0224a6c84d84","search","30c2722a-b901-11f0-b5de-0224a6c84d84",27,["D",1762207521254],[222],"31189d8a-b901-11f0-87ce-0224a6c84d84","data","31189e3e-b901-11f0-87ce-0224a6c84d84",29,["D",1762207521817],[228],"7eb3aa8e-b9ca-11f0-98dd-0224a6c84d84","answer","7eb3ab56-b9ca-11f0-98dd-0224a6c84d84",48,["D",1762293980863],[234],"pulls","branches","forks","envVars",{"_246":60,"_248":60},"connections",{"_246":60,"_247":60},"count","unfilledCount","emptyCount",{"_41":233,"_31":234,"_29":235,"_141":-5,"_39":40,"_38":236,"_143":47,"_36":431,"_13":60,"_33":34,"_35":-5,"_23":24,"_147":-5,"_279":432,"_26":433},{"_41":227,"_31":228,"_29":229,"_141":-5,"_39":40,"_38":230,"_143":47,"_36":428,"_13":60,"_33":34,"_35":-5,"_23":24,"_147":-5,"_279":429,"_26":430},{"_41":221,"_31":222,"_29":223,"_141":-5,"_39":40,"_38":224,"_143":47,"_36":425,"_13":60,"_33":34,"_35":-5,"_23":24,"_147":-5,"_279":426,"_26":427},{"_41":420,"_31":412,"_29":415,"_141":223,"_39":40,"_38":421,"_143":47,"_36":422,"_13":60,"_33":34,"_35":-5,"_23":24,"_147":-5,"_279":423,"_26":424},{"_41":414,"_31":413,"_29":408,"_141":415,"_39":40,"_38":416,"_143":47,"_36":417,"_13":60,"_33":34,"_35":-5,"_23":24,"_147":-5,"_279":418,"_26":419},{"_41":407,"_31":32,"_29":30,"_141":408,"_39":40,"_38":14,"_143":47,"_36":409,"_13":60,"_33":34,"_35":-5,"_23":24,"_147":-5,"_279":410,"_26":411},{"_41":399,"_31":400,"_29":401,"_141":223,"_39":40,"_38":402,"_143":24,"_36":403,"_13":60,"_33":145,"_35":404,"_23":24,"_147":-5,"_279":405,"_26":406},{"_41":391,"_31":392,"_29":393,"_141":223,"_39":40,"_38":394,"_143":47,"_36":395,"_13":60,"_33":145,"_35":396,"_23":24,"_147":-5,"_279":397,"_26":398},{"_41":383,"_31":384,"_29":385,"_141":223,"_39":40,"_38":386,"_143":47,"_36":387,"_13":60,"_33":145,"_35":388,"_23":24,"_147":-5,"_279":389,"_26":390},{"_41":375,"_31":376,"_29":377,"_141":223,"_39":40,"_38":378,"_143":47,"_36":379,"_13":60,"_33":145,"_35":380,"_23":24,"_147":-5,"_279":381,"_26":382},{"_41":367,"_31":368,"_29":369,"_141":223,"_39":40,"_38":370,"_143":47,"_36":371,"_13":60,"_33":145,"_35":372,"_23":24,"_147":-5,"_279":373,"_26":374},{"_41":359,"_31":360,"_29":361,"_141":223,"_39":40,"_38":362,"_143":24,"_36":363,"_13":60,"_33":145,"_35":364,"_23":24,"_147":-5,"_279":365,"_26":366},{"_41":351,"_31":352,"_29":353,"_141":223,"_39":40,"_38":354,"_143":47,"_36":355,"_13":60,"_33":145,"_35":356,"_23":24,"_147":-5,"_279":357,"_26":358},{"_41":344,"_31":139,"_29":345,"_141":223,"_39":40,"_38":346,"_143":24,"_36":347,"_13":60,"_33":145,"_35":348,"_23":24,"_147":-5,"_279":349,"_26":350},{"_41":215,"_31":216,"_29":217,"_141":-5,"_39":40,"_38":218,"_143":47,"_36":341,"_13":60,"_33":34,"_35":-5,"_23":24,"_147":-5,"_279":342,"_26":343},{"_41":333,"_31":334,"_29":335,"_141":217,"_39":40,"_38":336,"_143":24,"_36":337,"_13":60,"_33":145,"_35":338,"_23":24,"_147":-5,"_279":339,"_26":340},{"_41":325,"_31":326,"_29":327,"_141":217,"_39":40,"_38":328,"_143":24,"_36":329,"_13":60,"_33":145,"_35":330,"_23":24,"_147":-5,"_279":331,"_26":332},{"_41":317,"_31":318,"_29":319,"_141":217,"_39":40,"_38":320,"_143":47,"_36":321,"_13":60,"_33":145,"_35":322,"_23":24,"_147":-5,"_279":323,"_26":324},{"_41":309,"_31":310,"_29":311,"_141":217,"_39":40,"_38":312,"_143":24,"_36":313,"_13":60,"_33":145,"_35":314,"_23":24,"_147":-5,"_279":315,"_26":316},{"_41":209,"_31":210,"_29":211,"_141":-5,"_39":40,"_38":212,"_143":47,"_36":306,"_13":60,"_33":34,"_35":-5,"_23":24,"_147":-5,"_279":307,"_26":308},{"_41":202,"_31":203,"_29":204,"_141":-5,"_39":40,"_38":205,"_143":24,"_36":303,"_13":60,"_33":20,"_35":207,"_23":24,"_147":-5,"_279":304,"_26":305},{"_41":195,"_31":196,"_29":197,"_141":-5,"_39":40,"_38":198,"_143":47,"_36":300,"_13":60,"_33":20,"_35":200,"_23":24,"_147":-5,"_279":301,"_26":302},{"_41":188,"_31":189,"_29":190,"_141":-5,"_39":40,"_38":191,"_143":24,"_36":297,"_13":185,"_33":20,"_35":193,"_23":24,"_147":-5,"_279":298,"_26":299},{"_41":180,"_31":181,"_29":182,"_141":-5,"_39":40,"_38":183,"_143":24,"_36":294,"_13":185,"_33":20,"_35":186,"_23":24,"_147":-5,"_279":295,"_26":296},{"_41":173,"_31":174,"_29":175,"_141":-5,"_39":40,"_38":176,"_143":47,"_36":291,"_13":60,"_33":145,"_35":178,"_23":24,"_147":-5,"_279":292,"_26":293},{"_41":164,"_31":165,"_29":55,"_141":-5,"_39":40,"_38":166,"_143":24,"_36":288,"_13":168,"_33":169,"_35":170,"_23":24,"_147":171,"_279":289,"_26":290},{"_41":156,"_31":157,"_29":158,"_141":-5,"_39":40,"_38":159,"_143":24,"_36":285,"_13":161,"_33":20,"_35":162,"_23":24,"_147":-5,"_279":286,"_26":287},{"_41":149,"_31":150,"_29":151,"_141":-5,"_39":40,"_38":152,"_143":24,"_36":282,"_13":60,"_33":145,"_35":154,"_23":24,"_147":-5,"_279":283,"_26":284},{"_41":138,"_31":139,"_29":140,"_141":-5,"_39":40,"_38":142,"_143":47,"_36":278,"_13":60,"_33":145,"_35":146,"_23":24,"_147":-5,"_279":280,"_26":281},["D",1762200814128],"children",[],[139],["D",1762207522123],[],[150],["D",1762207522872],[],[157],["D",1762207522732],[],[165],["D",1762200814266],[],[174],["D",1762207522854],[],[181],["D",1762200814124],[],[189],["D",1762196442671],[],[196],["D",1762196442809],[],[203],["D",1762293981048],[],[210],"31b6d658-b901-11f0-bfbc-0224a6c84d84","test-search.ts","31b6d70c-b901-11f0-bfbc-0224a6c84d84",43,["D",1762207522693],"// Test script for search strategies\n// Run with: deno task test-search\n// Requires: Run `deno task recalculate` first to generate /data/latest.json\n\nimport \"jsr:@std/dotenv/load\";\nimport { getAllPagesForSearch } from \"../main.tsx\";\nimport { runSearchTest } from \"./search-harness.ts\";\nimport { questions } from \"./questions.ts\";\n\nasync function testSearchStrategies() {\n // Load cached data from latest.json\n let cachedData: any = null;\n try {\n const latestData = await Deno.readTextFile(\"./data/latest.json\");\n cachedData = JSON.parse(latestData);\n console.log(`Loaded cached data from /data/latest.json (${cachedData.totalPages} pages)`);\n } catch (error) {\n console.error(\"Failed to load /data/latest.json. Please run 'deno task recalculate' first.\");\n console.error(error);\n Deno.exit(1);\n }\n \n // Load actual pages with embeddings for search\n const pagesWithEmbeddings = await getAllPagesForSearch();\n console.log(`Loaded ${pagesWithEmbeddings.length} pages with embeddings for search`);\n \n // Load test questions\n const testQueries = questions;\n console.log(`Loaded ${testQueries.length} test questions\\n`);\n \n for (const query of testQueries) {\n console.log(`\\n${\"=\".repeat(60)}`);\n console.log(`Testing query: \"${query}\"`);\n console.log(`${\"=\".repeat(60)}`);\n \n const testResult = await runSearchTest(query, pagesWithEmbeddings, {\n limit: 10,\n minScore: 0,\n enableTiming: true, // Enable timing breakdown\n });\n \n if (testResult.error) {\n console.error(` ❌ Error: ${testResult.error}`);\n console.log(` ⏱️ Duration: ${testResult.durationMs.toFixed(2)}ms`);\n } else {\n console.log(` ✅ Found ${testResult.results.length} results in ${testResult.durationMs.toFixed(2)}ms\\n`);\n \n if (testResult.results.length === 0) {\n console.log(\" No results found.\");\n } else {\n testResult.results.forEach((result, index) => {\n console.log(` ${index + 1}. ${result.title || result.path}`);\n console.log(` Path: ${result.path}`);\n console.log(` Score: ${result.score.toFixed(2)}${result.metadata?.similarity ? ` (similarity: ${(result.metadata.similarity * 100).toFixed(2)}%)` : ''}`);\n if (result.snippet) {\n const snippet = result.snippet.length > 150 \n ? result.snippet.substring(0, 150) + '...' \n : result.snippet;\n console.log(` Snippet: ${snippet}`);\n }\n console.log('');\n });\n }\n }\n }\n \n console.log(`\\n${\"=\".repeat(60)}`);\n console.log(\"Test completed!\");\n console.log(`${\"=\".repeat(60)}\\n`);\n}\n\n// Run if executed directly\nif (import.meta.main) {\n testSearchStrategies().catch(console.error);\n}\n\n",[],[216,310],"3193ad5e-b901-11f0-a764-0224a6c84d84","search-harness.ts","3193ae26-b901-11f0-a764-0224a6c84d84",39,["D",1762207522627],"// Test harness for search functionality\n// Reusable function for testing search queries\n\nimport { searchPages } from \"../search/index.ts\";\nimport type { Page, SearchResult, SearchOptions } from \"../search/types.ts\";\n\nexport interface SearchTestResult {\n results: SearchResult[];\n durationMs: number;\n error?: string;\n}\n\n/**\n * Test harness function for running a single search query\n * @param query - The search query string\n * @param pages - Array of pages to search through\n * @param options - Search options (limit, minScore, etc.)\n * @returns Promise with results and timing information\n */\nexport async function runSearchTest(\n query: string,\n pages: Page[],\n options: SearchOptions = {}\n): Promise {\n const startTime = performance.now();\n \n try {\n const results = await searchPages(query, pages, options);\n const endTime = performance.now();\n const durationMs = endTime - startTime;\n \n return {\n results,\n durationMs,\n };\n } catch (error) {\n const endTime = performance.now();\n const durationMs = endTime - startTime;\n \n return {\n results: [],\n durationMs,\n error: error instanceof Error ? error.message : String(error),\n };\n }\n}\n\n",[],[216,318],"319c6fde-b901-11f0-a8ca-0224a6c84d84","recalculate.ts","319c70b0-b901-11f0-a8ca-0224a6c84d84",40,["D",1762207522645],"// Recalculate script - runs cache recalculation and saves to /data/latest.json and /data/mm-dd-yyyy.json\n// Run with: deno task recalculate\n\nimport \"jsr:@std/dotenv/load\";\nimport { urls } from \"../urls.ts\";\nimport {\n IS_VALTOWN,\n getFromCache,\n setCache,\n calculateContentHash,\n getTextFromUrl,\n parseFrontmatter,\n addUrlSourceToFrontmatter,\n calculateTokenCount,\n} from \"../utils.ts\";\nimport { generatePageMetadata } from \"../groq.ts\";\nimport { getActiveEmbeddingsFunction } from \"../search/index.ts\";\nimport { saveTestData } from \"../main.tsx\";\n\n// Load cached data from latest.json (for local hash checking)\nasync function loadLatestJson(): Promise | null> {\n if (IS_VALTOWN) {\n return null; // Use SQLite when in Val Town\n }\n \n try {\n const latestData = await Deno.readTextFile(\"./data/latest.json\");\n const data = JSON.parse(latestData);\n \n // Create a map of url -> cached data\n const cacheMap: Record = {};\n \n if (data.results && Array.isArray(data.results)) {\n for (const result of data.results) {\n if (result.url) {\n cacheMap[result.url] = {\n contentHash: result.contentHash,\n charCount: result.charCount,\n tokenCount: result.tokenCount,\n title: result.title,\n metadata: result.metadata,\n embeddings: result.embeddings || null,\n };\n }\n }\n }\n \n return cacheMap;\n } catch {\n // latest.json doesn't exist or is invalid - that's okay, just return null\n return null;\n }\n}\n\nasync function recalculate(force: boolean = false) {\n console.log(`Starting recalculation (force: ${force}, local: ${!IS_VALTOWN})...`);\n \n // Load latest.json for hash checking when running locally\n const latestJsonData = await loadLatestJson();\n if (latestJsonData && !force) {\n console.log(`Loaded ${Object.keys(latestJsonData).length} cached hashes from latest.json`);\n }\n \n const results: Array<{ \n path: string; \n url: string; \n charCount: number; \n tokenCount: number | undefined;\n title?: string;\n metadata?: any;\n contentHash?: string;\n embeddings?: number[] | null;\n error?: string;\n cacheError?: string;\n skipped?: boolean;\n reason?: string;\n }> = [];\n \n let processedCount = 0;\n let skippedCount = 0;\n \n const getPageByPath = async (pathParam: string, useCache: boolean = true) => {\n // Find the matching URL\n const matchingUrl = urls.find(url => {\n const match = url.match(/\\/docs\\/(.+?)\\.md$/);\n if (match) {\n return match[1] === pathParam;\n }\n if (url.endsWith(\"/docs.md\") && pathParam === \"docs\") {\n return true;\n }\n return false;\n });\n \n if (!matchingUrl) {\n return null;\n }\n \n // Try to get from cache first\n if (useCache) {\n const cached = await getFromCache(matchingUrl);\n if (cached) {\n return {\n url: cached.frontmatter?.urlSource || matchingUrl,\n content: cached.content,\n charCount: cached.charCount,\n tokenCount: cached.tokenCount,\n };\n }\n }\n \n // Fetch and process the single page\n const text = await getTextFromUrl(matchingUrl);\n const processedText = addUrlSourceToFrontmatter(text, matchingUrl);\n const charCount = processedText.length;\n const tokenCount = calculateTokenCount(processedText);\n \n return {\n url: matchingUrl,\n content: processedText,\n charCount,\n tokenCount,\n };\n };\n \n for (const url of urls) {\n // Extract path from URL\n const match = url.match(/\\/docs\\/(.+?)\\.md$/);\n const path = match ? match[1] : (url.endsWith(\"/docs.md\") ? \"docs\" : null);\n \n if (path) {\n console.log(`[${processedCount + 1}/${urls.length}] Processing ${path}...`);\n \n try {\n // Check cached hash if not forcing\n let shouldSkip = false;\n if (!force) {\n // Try SQLite cache first (if in Val Town)\n let cached = await getFromCache(url);\n \n // If no SQLite cache, try latest.json (for local development)\n let cachedHash: string | null = null;\n let cachedFromJson: any = null;\n \n if (cached) {\n cachedHash = cached.contentHash;\n } else if (latestJsonData && latestJsonData[url]) {\n cachedFromJson = latestJsonData[url];\n cachedHash = cachedFromJson.contentHash || null;\n }\n \n if (cachedHash) {\n // Fetch fresh content to check hash\n try {\n const pageData = await getPageByPath(path, false);\n if (pageData) {\n const newHash = await calculateContentHash(pageData.content);\n if (newHash === cachedHash) {\n shouldSkip = true;\n skippedCount++;\n console.log(` Skipped ${path} (content unchanged, hash: ${newHash.substring(0, 8)}...)`);\n \n // Use cached data from SQLite or latest.json\n if (cached) {\n results.push({\n path,\n url,\n charCount: cached.charCount,\n tokenCount: cached.tokenCount || undefined,\n title: (cached.frontmatter as any).title || path,\n metadata: cached.metadata || undefined,\n contentHash: cachedHash,\n embeddings: cached.embeddings || null,\n skipped: true,\n reason: \"Content unchanged (hash matches)\",\n });\n } else if (cachedFromJson) {\n results.push({\n path,\n url,\n charCount: cachedFromJson.charCount || 0,\n tokenCount: cachedFromJson.tokenCount || undefined,\n title: cachedFromJson.title || path,\n metadata: cachedFromJson.metadata || undefined,\n contentHash: cachedHash,\n embeddings: cachedFromJson.embeddings || null,\n skipped: true,\n reason: \"Content unchanged (hash matches)\",\n });\n }\n }\n }\n } catch (hashCheckError) {\n // If hash check fails, continue with processing\n console.debug(` Hash check failed for ${path}, continuing...`);\n }\n }\n }\n \n if (shouldSkip) {\n processedCount++;\n continue;\n }\n \n // Fetch without cache\n let pageData;\n try {\n pageData = await getPageByPath(path, false);\n } catch (fetchError) {\n const fetchErrorMessage = fetchError instanceof Error ? fetchError.message : String(fetchError);\n console.error(` Failed to fetch ${path}:`, fetchErrorMessage);\n throw new Error(`Fetch failed: ${fetchErrorMessage}`);\n }\n \n if (pageData) {\n const frontmatter = parseFrontmatter(pageData.content);\n const title = (frontmatter as any).title || path;\n \n // Calculate content hash\n const contentHash = await calculateContentHash(pageData.content);\n \n // Generate AI metadata\n const metadata = await generatePageMetadata(pageData.content, title);\n if (!metadata) {\n console.warn(` Metadata generation failed for ${path} (continuing without metadata)`);\n }\n \n // Generate embeddings using active strategy's embeddings function\n const generateEmbeddings = getActiveEmbeddingsFunction();\n let embeddings: number[] | null = null;\n let embeddingsError: string | undefined = undefined;\n \n try {\n embeddings = await generateEmbeddings(pageData.content);\n if (!embeddings) {\n embeddingsError = \"Embeddings generation returned null\";\n console.error(` ❌ Embeddings generation failed for ${path}: ${embeddingsError}`);\n }\n } catch (embedError) {\n const embedErrorMessage = embedError instanceof Error ? embedError.message : String(embedError);\n embeddingsError = `Embeddings generation threw error: ${embedErrorMessage}`;\n console.error(` ❌ Embeddings generation failed for ${path}: ${embeddingsError}`);\n // Don't throw - continue with null embeddings, but log the error\n }\n \n // Cache it with metadata, hash, and embeddings - track errors\n const cacheError = await setCache(url, {\n content: pageData.content,\n charCount: pageData.charCount,\n tokenCount: pageData.tokenCount || null,\n frontmatter,\n metadata,\n contentHash,\n embeddings,\n });\n \n results.push({ \n path, \n url, \n charCount: pageData.charCount, \n tokenCount: pageData.tokenCount,\n title,\n metadata,\n contentHash,\n embeddings: embeddings || null,\n ...(cacheError && { cacheError }),\n ...(embeddingsError && { embeddingsError }),\n });\n \n const statusParts = [\n `metadata: ${metadata ? 'yes' : 'no'}`,\n `embeddings: ${embeddings ? 'yes' : 'no'}`,\n ];\n \n if (embeddingsError) {\n statusParts.push(`embeddings error: ${embeddingsError.substring(0, 50)}...`);\n }\n if (cacheError) {\n statusParts.push(`cache error: ${cacheError}`);\n }\n \n const status = statusParts.join(', ');\n if (embeddingsError || cacheError) {\n console.warn(`Completed ${path} (${status})`);\n } else {\n console.log(`Completed ${path} (${status}, cached: yes)`);\n }\n }\n } catch (error) {\n const errorMessage = error instanceof Error ? error.message : String(error);\n console.error(`Error processing ${path}:`, errorMessage);\n results.push({\n path,\n url,\n charCount: 0,\n tokenCount: 0,\n error: errorMessage,\n });\n }\n \n processedCount++;\n }\n }\n \n const cacheErrors = results.filter(r => r.cacheError);\n const processedResults = results.filter(r => !r.skipped);\n \n const summary = {\n message: force \n ? `Recalculated ${processedResults.length} pages with AI metadata (force mode)` \n : `Recalculated ${processedResults.length} pages, skipped ${skippedCount} unchanged pages`,\n timestamp: new Date().toISOString(),\n results,\n totalPages: results.length,\n processed: processedResults.length,\n skipped: skippedCount,\n withMetadata: processedResults.filter(r => r.metadata).length,\n withoutMetadata: processedResults.filter(r => !r.metadata).length,\n cached: processedResults.length > 0 && cacheErrors.length === 0,\n cacheErrors: cacheErrors.length > 0 ? cacheErrors.map(r => ({ path: r.path, error: r.cacheError })) : undefined,\n force,\n };\n \n // Save to both latest.json and dated file\n await saveTestData(summary);\n console.log(\"\\nRecalculation completed!\");\n \n return summary;\n}\n\n// Run if executed directly\nif (import.meta.main) {\n const force = Deno.args.includes(\"--force\") || Deno.args.includes(\"-f\");\n recalculate(force).catch(console.error);\n}\n\n",[],[216,326],"31a38e68-b901-11f0-a3b9-0224a6c84d84","questions.ts","31a38f30-b901-11f0-a3b9-0224a6c84d84",41,["D",1762207522633],"// Test questions for search testing\n// Full sentence questions about Groq\n\nexport const questions: string[] = [\n \"What is Compound and how does it work?\",\n // \"How do I authenticate with the Groq API?\",\n // \"What are the rate limits for Groq models?\",\n // \"How do I integrate Groq with my JavaScript application?\",\n // \"What is the difference between compound and compound-mini?\",\n // \"Can I use Groq for real-time web search?\",\n // \"How do I get started with Groq's API?\",\n // \"What models are available on Groq?\",\n // \"How does Groq's Compound AI system work?\",\n // \"Can I use custom tools with Groq's Compound AI?\",\n];\n\n",[],[216,334],["D",1762207521420],[264,265,266,267],[216],"31528cd4-b901-11f0-a3b9-0224a6c84d84","31528d88-b901-11f0-a3b9-0224a6c84d84",32,["D",1762207522155],"// Shared utilities for search strategies\n\n// Calculate cosine similarity between two embedding vectors\nexport const cosineSimilarity = (vecA: number[], vecB: number[]): number => {\n if (vecA.length !== vecB.length) {\n return 0;\n }\n \n let dotProduct = 0;\n let normA = 0;\n let normB = 0;\n \n for (let i = 0; i < vecA.length; i++) {\n dotProduct += vecA[i] * vecB[i];\n normA += vecA[i] * vecA[i];\n normB += vecB[i] * vecB[i];\n }\n \n if (normA === 0 || normB === 0) {\n return 0;\n }\n \n return dotProduct / (Math.sqrt(normA) * Math.sqrt(normB));\n};\n\n// Helper function to generate snippets\nexport const generateSnippet = (content: string, queryWords: string[], queryLower: string): string => {\n const contentLower = content.toLowerCase();\n const firstMatchIndex = contentLower.indexOf(queryWords[0]);\n \n if (firstMatchIndex >= 0) {\n const start = Math.max(0, firstMatchIndex - 50);\n const end = Math.min(content.length, firstMatchIndex + queryWords[0].length + 150);\n let snippet = content.substring(start, end);\n if (start > 0) snippet = '...' + snippet;\n if (end < content.length) snippet = snippet + '...';\n return snippet;\n }\n \n return content.substring(0, 200) + '...';\n};\n\n",[],[222,139],"316a968a-b901-11f0-9996-0224a6c84d84","types.ts","316a973e-b901-11f0-9996-0224a6c84d84",34,["D",1762207522301],"// Search strategy types and interfaces\n\nexport type SearchResult = {\n path: string;\n url: string;\n title?: string;\n score: number;\n snippet?: string;\n metadata?: {\n strategy?: string;\n [key: string]: any;\n };\n};\n\nexport type Page = {\n path: string;\n url: string;\n content: string;\n title?: string;\n embeddings?: number[] | null;\n metadata?: any;\n};\n\nexport type SearchOptions = {\n limit?: number;\n minScore?: number;\n [key: string]: any;\n};\n\nexport type SearchStrategy = {\n name: string;\n description: string;\n search: (query: string, pages: Page[], options: SearchOptions) => Promise;\n};\n\n",[],[222,352],"31884234-b901-11f0-b784-0224a6c84d84","transformers-cosine.ts","31884324-b901-11f0-b784-0224a6c84d84",38,["D",1762207522387],"// Transformers.js + Cosine Similarity Strategy: Local embeddings in browser/Deno\n// Fastest option - no API calls, runs entirely client-side\n// Model: all-MiniLM-L6-v2 (80MB, ~384 dimensions, very fast)\n\nimport type { SearchStrategy, SearchResult, Page, SearchOptions } from \"./types.ts\";\nimport { generateSnippet, cosineSimilarity } from \"./utils.ts\";\n\n// Lazy-loaded pipeline (cached after first load)\nlet embeddingPipeline: any = null;\nlet pipelineLoading: Promise | null = null;\n\nasync function getEmbeddingPipeline() {\n // Return cached pipeline if already loaded\n if (embeddingPipeline) {\n return embeddingPipeline;\n }\n \n // If loading in progress, wait for it\n if (pipelineLoading) {\n return pipelineLoading;\n }\n \n // Start loading pipeline\n pipelineLoading = (async () => {\n try {\n // Import transformers.js - use npm: prefix for Deno compatibility\n // NOTE: transformers.js is primarily designed for browsers. In Deno, it may require --allow-ffi\n // For browser usage, this will work seamlessly. For Deno, you may need additional permissions.\n const { pipeline } = await import(\"npm:@huggingface/transformers@latest\");\n \n // Load feature extraction pipeline with all-MiniLM-L6-v2 model\n // This is a small, fast model optimized for embeddings (80MB, ~384 dims)\n embeddingPipeline = await pipeline(\n 'feature-extraction',\n 'Xenova/all-MiniLM-L6-v2',\n {\n // Use CPU - WebGPU might not be available in all environments\n device: 'cpu',\n dtype: 'fp32',\n }\n );\n \n return embeddingPipeline;\n } catch (error) {\n const errorMessage = error instanceof Error ? error.message : String(error);\n const fullError = `Failed to load transformers.js pipeline: ${errorMessage}\\n` +\n `NOTE: transformers.js is primarily designed for browser environments. ` +\n `If running in Deno, you may need --allow-ffi permissions, or use a different embedding strategy.`;\n console.error(fullError);\n throw new Error(fullError);\n } finally {\n pipelineLoading = null;\n }\n })();\n \n return pipelineLoading;\n}\n\n// Generate embeddings using transformers.js (runs locally, no API calls)\nexport const generateEmbeddings = async (content: string): Promise => {\n try {\n const pipeline = await getEmbeddingPipeline();\n \n // Generate embeddings - returns a tensor, we need to extract the array\n const output = await pipeline(content, {\n pooling: 'mean',\n normalize: true,\n });\n \n // Extract the embedding vector from the tensor\n // Transformers.js returns a Tensor object - handle different formats\n if (output && typeof output === 'object') {\n // Check for .data property (common tensor format)\n if (output.data && Array.isArray(output.data)) {\n return output.data;\n }\n \n // Check if it's a nested array (tensor shape)\n if (Array.isArray(output)) {\n // Flatten if nested\n if (Array.isArray(output[0])) {\n return output[0]; // Return first row if 2D\n }\n return output;\n }\n \n // Try accessing tensor properties\n if (output.size && output.data) {\n return Array.from(output.data);\n }\n \n // Try tolist() method if available\n if (typeof output.tolist === 'function') {\n const list = output.tolist();\n return Array.isArray(list[0]) ? list[0] : list;\n }\n }\n \n // Fallback: try to convert to array\n return Array.from(output as any);\n } catch (error) {\n const errorMessage = error instanceof Error ? error.message : String(error);\n console.error(\"Transformers.js embeddings failed:\", errorMessage);\n return null;\n }\n};\n\nexport const searchStrategy: SearchStrategy = {\n name: \"transformers-cosine\",\n description: \"Semantic search using Transformers.js local embeddings with cosine similarity (fastest, no API calls)\",\n search: async (query: string, pages: Page[], options: SearchOptions = {}): Promise => {\n const limit = options.limit || 10;\n const minSimilarity = (options.minScore || 0) / 100; // Convert to 0-1 scale\n const enableTiming = options.enableTiming || false;\n \n const timings: Record = {};\n const startTime = performance.now();\n \n // Generate query embedding (local, no API call)\n const queryEmbedStart = performance.now();\n const queryEmbedding = await generateEmbeddings(query);\n if (enableTiming) {\n timings.queryEmbedding = performance.now() - queryEmbedStart;\n }\n \n if (!queryEmbedding) {\n console.warn(\"Query embedding generation failed, returning empty results\");\n return [];\n }\n \n // Calculate cosine similarity for each page\n const similarityStart = performance.now();\n const scoredPages: Array<{ page: Page; similarity: number }> = [];\n \n for (const page of pages) {\n const pageEmbedding = page.embeddings;\n if (!pageEmbedding || pageEmbedding.length !== queryEmbedding.length) {\n continue; // Skip pages without embeddings or wrong dimension\n }\n \n const similarity = cosineSimilarity(queryEmbedding, pageEmbedding);\n if (similarity >= minSimilarity) {\n scoredPages.push({ page, similarity });\n }\n }\n \n if (enableTiming) timings.similarity = performance.now() - similarityStart;\n \n // Sort by similarity (descending) and limit\n const sortStart = performance.now();\n scoredPages.sort((a, b) => b.similarity - a.similarity);\n const topPages = scoredPages.slice(0, limit);\n if (enableTiming) timings.sort = performance.now() - sortStart;\n \n // Transform to SearchResult format\n const transformStart = performance.now();\n const queryWords = query.toLowerCase().split(/\\s+/).filter(w => w.length > 0);\n const results: SearchResult[] = topPages.map(({ page, similarity }) => ({\n path: page.path,\n url: page.url,\n title: page.title || page.path,\n score: similarity * 100, // Convert to 0-100 scale\n snippet: generateSnippet(page.content, queryWords, query.toLowerCase()),\n metadata: {\n strategy: \"transformers-cosine\",\n similarity,\n ...(enableTiming && { timings }),\n },\n }));\n if (enableTiming) timings.transform = performance.now() - transformStart;\n \n if (enableTiming) {\n timings.total = performance.now() - startTime;\n console.log(` ⏱️ Timing breakdown:`);\n console.log(` Query embedding: ${timings.queryEmbedding?.toFixed(2)}ms (local model - no API)`);\n console.log(` Cosine similarity: ${timings.similarity?.toFixed(2)}ms`);\n console.log(` Sort: ${timings.sort?.toFixed(2)}ms`);\n console.log(` Transform: ${timings.transform?.toFixed(2)}ms`);\n console.log(` Total: ${timings.total?.toFixed(2)}ms`);\n }\n \n return results;\n },\n};\n\n",[],[222,360],"3181e164-b901-11f0-a3b9-0224a6c84d84","placeholder.ts","3181e22c-b901-11f0-a3b9-0224a6c84d84",37,["D",1762207522458],"// Placeholder Strategy: Keyword-based search (current default)\n\nimport type { SearchStrategy, SearchResult, Page, SearchOptions } from \"./types.ts\";\nimport { generateSnippet } from \"./utils.ts\";\n\n// Fake embeddings function for placeholder (not actually used, but kept for interface consistency)\nexport const generateEmbeddings = async (content: string): Promise => {\n // Not used in placeholder strategy, but kept for consistency\n // Returns fake embeddings for caching purposes\n console.debug(\"Generating fake embeddings for placeholder strategy\");\n \n const fakeEmbedding: number[] = [];\n const hash = content.split('').reduce((acc, char) => acc + char.charCodeAt(0), 0);\n \n for (let i = 0; i < 384; i++) {\n const value = Math.sin(hash + i) * 0.5 + 0.5;\n fakeEmbedding.push(value);\n }\n \n return fakeEmbedding;\n};\n\nexport const searchStrategy: SearchStrategy = {\n name: \"placeholder\",\n description: \"Keyword-based search using title, content, and metadata matching\",\n search: async (query: string, pages: Page[], options: SearchOptions = {}): Promise => {\n const limit = options.limit || 10;\n const minScore = options.minScore || 0;\n \n const queryLower = query.toLowerCase();\n const queryWords = queryLower.split(/\\s+/).filter(w => w.length > 0);\n \n const results: SearchResult[] = [];\n \n for (const page of pages) {\n let score = 0;\n const contentLower = page.content.toLowerCase();\n const titleLower = (page.title || page.path).toLowerCase();\n \n // Title matches are worth more\n for (const word of queryWords) {\n if (titleLower.includes(word)) {\n score += 10;\n }\n // Count occurrences in content\n const contentMatches = (contentLower.match(new RegExp(word, 'g')) || []).length;\n score += Math.min(contentMatches, 5); // Cap at 5 points per word\n }\n \n // Exact phrase match bonus\n if (contentLower.includes(queryLower)) {\n score += 20;\n }\n \n // Metadata boost (if query matches tags or categories)\n if (page.metadata) {\n const tags = (page.metadata.tags || []).join(' ').toLowerCase();\n const categories = (page.metadata.categories || []).join(' ').toLowerCase();\n const useCases = (page.metadata.useCases || []).join(' ').toLowerCase();\n \n for (const word of queryWords) {\n if (tags.includes(word)) score += 3;\n if (categories.includes(word)) score += 2;\n if (useCases.includes(word)) score += 2;\n }\n }\n \n if (score > minScore) {\n results.push({\n path: page.path,\n url: page.url,\n title: page.title,\n score,\n snippet: generateSnippet(page.content, queryWords, queryLower),\n metadata: {\n strategy: \"placeholder\",\n },\n });\n }\n }\n \n // Sort by score descending\n results.sort((a, b) => b.score - a.score);\n \n return results.slice(0, limit);\n },\n};\n\n",[],[222,368],"314db04c-b901-11f0-a764-0224a6c84d84","openai-orama.ts","314db0ec-b901-11f0-a764-0224a6c84d84",31,["D",1762207522111],"// OpenAI + Orama Strategy: Semantic search using OpenAI embeddings and Orama\n// Faster than JigsawStack (~100-200ms vs ~550ms for query embeddings)\n\nimport type { SearchStrategy, SearchResult, Page, SearchOptions } from \"./types.ts\";\nimport { generateSnippet } from \"./utils.ts\";\nimport { create, insertMultiple, search } from \"npm:@orama/orama@latest\";\n\n// OpenAI embeddings function\nexport const generateEmbeddings = async (content: string): Promise => {\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n const OPENAI_API_URL = \"https://api.openai.com/v1/embeddings\";\n \n if (!OPENAI_API_KEY) {\n console.warn(\"OPENAI_API_KEY not found - embeddings disabled\");\n return null;\n }\n \n try {\n const response = await fetch(OPENAI_API_URL, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n model: \"text-embedding-3-small\", // Fast and efficient model\n input: content,\n }),\n });\n \n if (!response.ok) {\n const errorText = await response.text().catch(() => response.statusText);\n throw new Error(`OpenAI API error: ${response.status} ${response.statusText} - ${errorText.substring(0, 200)}`);\n }\n \n const data = await response.json();\n \n if (data.data && Array.isArray(data.data) && data.data.length > 0) {\n return data.data[0].embedding; // Return the embedding vector\n }\n \n return null;\n } catch (error) {\n const errorMessage = error instanceof Error ? error.message : String(error);\n console.error(\"OpenAI embeddings failed:\", errorMessage);\n return null;\n }\n};\n\n// No caching for query embeddings - we want to benchmark actual API performance\n\nexport const searchStrategy: SearchStrategy = {\n name: \"openai-orama\",\n description: \"Semantic search using OpenAI embeddings with Orama vector search (faster than JigsawStack)\",\n search: async (query: string, pages: Page[], options: SearchOptions = {}): Promise => {\n const limit = options.limit || 10;\n const minSimilarity = (options.minScore || 0) / 100; // Convert to 0-1 scale\n const enableTiming = options.enableTiming || false;\n \n const timings: Record = {};\n const startTime = performance.now();\n \n // Generate query embedding (no caching - benchmarking actual API performance)\n const queryEmbedStart = performance.now();\n const queryEmbedding = await generateEmbeddings(query);\n if (enableTiming) {\n timings.queryEmbedding = performance.now() - queryEmbedStart;\n }\n \n if (!queryEmbedding) {\n console.warn(\"Query embedding generation failed, returning empty results\");\n return [];\n }\n \n // Determine embedding dimension from query (or first page with embeddings)\n const embeddingDimension = queryEmbedding.length;\n if (embeddingDimension === 0) {\n return [];\n }\n \n // Create Orama database with vector field\n const dbCreateStart = performance.now();\n const db = await create({\n schema: {\n path: \"string\",\n url: \"string\",\n title: \"string\",\n content: \"string\",\n embedding: `vector[${embeddingDimension}]`,\n },\n });\n if (enableTiming) timings.dbCreate = performance.now() - dbCreateStart;\n \n // Prepare documents with embeddings for insertion\n const documents: Array<{\n path: string;\n url: string;\n title: string;\n content: string;\n embedding: number[];\n }> = [];\n \n // Collect pages with embeddings\n const docPrepStart = performance.now();\n for (const page of pages) {\n // Use cached embeddings if available, otherwise generate\n let pageEmbedding = page.embeddings;\n if (!pageEmbedding) {\n // Generate embedding if not cached (should be cached from recalculate, but handle missing)\n pageEmbedding = await generateEmbeddings(page.content);\n if (!pageEmbedding || pageEmbedding.length !== embeddingDimension) {\n continue; // Skip pages without embeddings or wrong dimension\n }\n }\n \n // Ensure embedding dimension matches\n if (pageEmbedding.length !== embeddingDimension) {\n continue;\n }\n \n documents.push({\n path: page.path,\n url: page.url,\n title: page.title || page.path,\n content: page.content,\n embedding: pageEmbedding,\n });\n }\n if (enableTiming) timings.docPrep = performance.now() - docPrepStart;\n \n // Insert all documents into Orama\n const insertStart = performance.now();\n if (documents.length > 0) {\n await insertMultiple(db, documents);\n } else {\n return []; // No documents to search\n }\n if (enableTiming) timings.insert = performance.now() - insertStart;\n \n // Perform vector similarity search using Orama\n const searchStart = performance.now();\n const searchResults = await search(db, {\n mode: \"vector\",\n vector: {\n value: queryEmbedding,\n property: \"embedding\",\n },\n similarity: minSimilarity,\n limit,\n includeVectors: false, // Don't need vectors in response\n });\n if (enableTiming) timings.search = performance.now() - searchStart;\n \n // Transform Orama results to SearchResult format\n const transformStart = performance.now();\n const results: SearchResult[] = searchResults.hits.map((hit: any) => {\n const page = pages.find(p => p.path === hit.document.path);\n const queryWords = query.toLowerCase().split(/\\s+/).filter(w => w.length > 0);\n \n return {\n path: hit.document.path,\n url: hit.document.url,\n title: hit.document.title,\n score: (hit.score || 0) * 100, // Convert similarity to score (0-100)\n snippet: page ? generateSnippet(page.content, queryWords, query.toLowerCase()) : hit.document.content.substring(0, 200) + '...',\n metadata: {\n strategy: \"openai-orama\",\n similarity: hit.score,\n ...(enableTiming && { timings }),\n },\n };\n });\n if (enableTiming) timings.transform = performance.now() - transformStart;\n \n if (enableTiming) {\n timings.total = performance.now() - startTime;\n console.log(` ⏱️ Timing breakdown:`);\n console.log(` Query embedding: ${timings.queryEmbedding?.toFixed(2)}ms (API call)`);\n console.log(` DB create: ${timings.dbCreate?.toFixed(2)}ms`);\n console.log(` Doc prep: ${timings.docPrep?.toFixed(2)}ms`);\n console.log(` Insert: ${timings.insert?.toFixed(2)}ms`);\n console.log(` Search: ${timings.search?.toFixed(2)}ms`);\n console.log(` Transform: ${timings.transform?.toFixed(2)}ms`);\n console.log(` Total: ${timings.total?.toFixed(2)}ms`);\n }\n \n return results;\n },\n};\n\n",[],[222,376],"31459b96-b901-11f0-9996-0224a6c84d84","openai-cosine.ts","31459c22-b901-11f0-9996-0224a6c84d84",30,["D",1762207522106],"// OpenAI + Cosine Similarity Strategy: Direct cosine similarity calculation\n// Fastest for small datasets (<100 pages) - no DB overhead\n\nimport type { SearchStrategy, SearchResult, Page, SearchOptions } from \"./types.ts\";\nimport { generateSnippet, cosineSimilarity } from \"./utils.ts\";\n\n// OpenAI embeddings function\nexport const generateEmbeddings = async (content: string): Promise => {\n const OPENAI_API_KEY = Deno.env.get(\"OPENAI_API_KEY\");\n const OPENAI_API_URL = \"https://api.openai.com/v1/embeddings\";\n \n if (!OPENAI_API_KEY) {\n console.warn(\"OPENAI_API_KEY not found - embeddings disabled\");\n return null;\n }\n \n try {\n const response = await fetch(OPENAI_API_URL, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${OPENAI_API_KEY}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n model: \"text-embedding-3-small\", // Fast and efficient model\n input: content,\n }),\n });\n \n if (!response.ok) {\n const errorText = await response.text().catch(() => response.statusText);\n throw new Error(`OpenAI API error: ${response.status} ${response.statusText} - ${errorText.substring(0, 200)}`);\n }\n \n const data = await response.json();\n \n if (data.data && Array.isArray(data.data) && data.data.length > 0) {\n return data.data[0].embedding; // Return the embedding vector\n }\n \n return null;\n } catch (error) {\n const errorMessage = error instanceof Error ? error.message : String(error);\n console.error(\"OpenAI embeddings failed:\", errorMessage);\n return null;\n }\n};\n\n// No caching for query embeddings - we want to benchmark actual API performance\n\n\nexport const searchStrategy: SearchStrategy = {\n name: \"openai-cosine\",\n description: \"Semantic search using OpenAI embeddings with direct cosine similarity (fastest for small datasets)\",\n search: async (query: string, pages: Page[], options: SearchOptions = {}): Promise => {\n const limit = options.limit || 10;\n const minSimilarity = (options.minScore || 0) / 100; // Convert to 0-1 scale\n const enableTiming = options.enableTiming || false;\n \n const timings: Record = {};\n const startTime = performance.now();\n \n // Generate query embedding (no caching - benchmarking actual API performance)\n const queryEmbedStart = performance.now();\n const queryEmbedding = await generateEmbeddings(query);\n if (enableTiming) {\n timings.queryEmbedding = performance.now() - queryEmbedStart;\n }\n \n if (!queryEmbedding) {\n console.warn(\"Query embedding generation failed, returning empty results\");\n return [];\n }\n \n // Calculate cosine similarity for each page\n const similarityStart = performance.now();\n const scoredPages: Array<{ page: Page; similarity: number }> = [];\n \n for (const page of pages) {\n const pageEmbedding = page.embeddings;\n if (!pageEmbedding || pageEmbedding.length !== queryEmbedding.length) {\n continue; // Skip pages without embeddings or wrong dimension\n }\n \n const similarity = cosineSimilarity(queryEmbedding, pageEmbedding);\n if (similarity >= minSimilarity) {\n scoredPages.push({ page, similarity });\n }\n }\n \n if (enableTiming) timings.similarity = performance.now() - similarityStart;\n \n // Sort by similarity (descending) and limit\n const sortStart = performance.now();\n scoredPages.sort((a, b) => b.similarity - a.similarity);\n const topPages = scoredPages.slice(0, limit);\n if (enableTiming) timings.sort = performance.now() - sortStart;\n \n // Transform to SearchResult format\n const transformStart = performance.now();\n const queryWords = query.toLowerCase().split(/\\s+/).filter(w => w.length > 0);\n const results: SearchResult[] = topPages.map(({ page, similarity }) => ({\n path: page.path,\n url: page.url,\n title: page.title || page.path,\n score: similarity * 100, // Convert to 0-100 scale\n snippet: generateSnippet(page.content, queryWords, query.toLowerCase()),\n metadata: {\n strategy: \"openai-cosine\",\n similarity,\n ...(enableTiming && { timings }),\n },\n }));\n if (enableTiming) timings.transform = performance.now() - transformStart;\n \n if (enableTiming) {\n timings.total = performance.now() - startTime;\n console.log(` ⏱️ Timing breakdown:`);\n console.log(` Query embedding: ${timings.queryEmbedding?.toFixed(2)}ms (API call)`);\n console.log(` Cosine similarity: ${timings.similarity?.toFixed(2)}ms`);\n console.log(` Sort: ${timings.sort?.toFixed(2)}ms`);\n console.log(` Transform: ${timings.transform?.toFixed(2)}ms`);\n console.log(` Total: ${timings.total?.toFixed(2)}ms`);\n }\n \n return results;\n },\n};\n\n",[],[222,384],"315f55ae-b901-11f0-8f9e-0224a6c84d84","jigsawstack-orama.ts","315f56bc-b901-11f0-8f9e-0224a6c84d84",33,["D",1762207522117],"// JigsawStack + Orama Strategy: Semantic search using JigsawStack embeddings and Orama\n\nimport type { SearchStrategy, SearchResult, Page, SearchOptions } from \"./types.ts\";\nimport { generateSnippet } from \"./utils.ts\";\nimport { create, insertMultiple, search } from \"npm:@orama/orama@latest\";\n\n// JigsawStack embeddings function (v2 API)\nexport const generateEmbeddings = async (content: string): Promise => {\n const JIGSAWSTACK_API_KEY = Deno.env.get(\"JIGSAWSTACK_API_KEY\");\n const JIGSAWSTACK_API_URL = \"https://api.jigsawstack.com/v2/embedding\";\n \n if (!JIGSAWSTACK_API_KEY) {\n console.warn(\"JIGSAWSTACK_API_KEY not found - embeddings disabled\");\n return null;\n }\n \n try {\n const response = await fetch(JIGSAWSTACK_API_URL, {\n method: \"POST\",\n headers: {\n \"X-Api-Key\": JIGSAWSTACK_API_KEY,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n text: content,\n type: \"text\",\n }),\n });\n \n if (!response.ok) {\n const errorText = await response.text().catch(() => response.statusText);\n throw new Error(`JigsawStack API error: ${response.status} ${response.statusText} - ${errorText.substring(0, 200)}`);\n }\n \n const data = await response.json();\n \n // JigsawStack v2 returns: { success: true, embeddings: [[...]], chunks: [...], _usage: {...} }\n // embeddings is an array of arrays, we want the first embedding vector\n if (data.success && data.embeddings && Array.isArray(data.embeddings) && data.embeddings.length > 0) {\n return data.embeddings[0]; // Return the first embedding vector\n }\n \n return null;\n } catch (error) {\n const errorMessage = error instanceof Error ? error.message : String(error);\n console.error(\"JigsawStack embeddings failed:\", errorMessage);\n return null;\n }\n};\n\nexport const searchStrategy: SearchStrategy = {\n name: \"jigsawstack-orama\",\n description: \"Semantic search using JigsawStack embeddings with Orama vector search\",\n search: async (query: string, pages: Page[], options: SearchOptions = {}): Promise => {\n const limit = options.limit || 10;\n const minSimilarity = (options.minScore || 0) / 100; // Convert to 0-1 scale\n const enableTiming = options.enableTiming || false;\n \n const timings: Record = {};\n const startTime = performance.now();\n \n // Generate query embedding (this is the main bottleneck - external API call)\n const queryEmbedStart = performance.now();\n const queryEmbedding = await generateEmbeddings(query);\n if (enableTiming) timings.queryEmbedding = performance.now() - queryEmbedStart;\n \n if (!queryEmbedding) {\n console.warn(\"Query embedding generation failed, returning empty results\");\n return [];\n }\n \n // Determine embedding dimension from query (or first page with embeddings)\n const embeddingDimension = queryEmbedding.length;\n if (embeddingDimension === 0) {\n return [];\n }\n \n // Create Orama database with vector field\n const dbCreateStart = performance.now();\n const db = await create({\n schema: {\n path: \"string\",\n url: \"string\",\n title: \"string\",\n content: \"string\",\n embedding: `vector[${embeddingDimension}]`,\n },\n });\n if (enableTiming) timings.dbCreate = performance.now() - dbCreateStart;\n \n // Prepare documents with embeddings for insertion\n const documents: Array<{\n path: string;\n url: string;\n title: string;\n content: string;\n embedding: number[];\n }> = [];\n \n // Collect pages with embeddings\n const docPrepStart = performance.now();\n for (const page of pages) {\n // Use cached embeddings if available, otherwise generate\n let pageEmbedding = page.embeddings;\n if (!pageEmbedding) {\n // Generate embedding if not cached (should be cached from recalculate, but handle missing)\n pageEmbedding = await generateEmbeddings(page.content);\n if (!pageEmbedding || pageEmbedding.length !== embeddingDimension) {\n continue; // Skip pages without embeddings or wrong dimension\n }\n }\n \n // Ensure embedding dimension matches\n if (pageEmbedding.length !== embeddingDimension) {\n continue;\n }\n \n documents.push({\n path: page.path,\n url: page.url,\n title: page.title || page.path,\n content: page.content,\n embedding: pageEmbedding,\n });\n }\n if (enableTiming) timings.docPrep = performance.now() - docPrepStart;\n \n // Insert all documents into Orama\n const insertStart = performance.now();\n if (documents.length > 0) {\n await insertMultiple(db, documents);\n } else {\n return []; // No documents to search\n }\n if (enableTiming) timings.insert = performance.now() - insertStart;\n \n // Perform vector similarity search using Orama\n const searchStart = performance.now();\n const searchResults = await search(db, {\n mode: \"vector\",\n vector: {\n value: queryEmbedding,\n property: \"embedding\",\n },\n similarity: minSimilarity,\n limit,\n includeVectors: false, // Don't need vectors in response\n });\n if (enableTiming) timings.search = performance.now() - searchStart;\n \n // Transform Orama results to SearchResult format\n const transformStart = performance.now();\n const results: SearchResult[] = searchResults.hits.map((hit: any) => {\n const page = pages.find(p => p.path === hit.document.path);\n const queryWords = query.toLowerCase().split(/\\s+/).filter(w => w.length > 0);\n \n return {\n path: hit.document.path,\n url: hit.document.url,\n title: hit.document.title,\n score: (hit.score || 0) * 100, // Convert similarity to score (0-100)\n snippet: page ? generateSnippet(page.content, queryWords, query.toLowerCase()) : hit.document.content.substring(0, 200) + '...',\n metadata: {\n strategy: \"jigsawstack-orama\",\n similarity: hit.score,\n ...(enableTiming && { timings }),\n },\n };\n });\n if (enableTiming) timings.transform = performance.now() - transformStart;\n \n if (enableTiming) {\n timings.total = performance.now() - startTime;\n console.log(` ⏱️ Timing breakdown:`);\n console.log(` Query embedding: ${timings.queryEmbedding?.toFixed(2)}ms (API call - main bottleneck)`);\n console.log(` DB create: ${timings.dbCreate?.toFixed(2)}ms`);\n console.log(` Doc prep: ${timings.docPrep?.toFixed(2)}ms`);\n console.log(` Insert: ${timings.insert?.toFixed(2)}ms`);\n console.log(` Search: ${timings.search?.toFixed(2)}ms`);\n console.log(` Transform: ${timings.transform?.toFixed(2)}ms`);\n console.log(` Total: ${timings.total?.toFixed(2)}ms`);\n }\n \n return results;\n },\n};\n\n",[],[222,392],"31712298-b901-11f0-878c-0224a6c84d84","index.ts","3171232e-b901-11f0-878c-0224a6c84d84",35,["D",1762207522385],"// Search - Main entry point\n// To switch strategies, comment/uncomment the imports and function calls below\n\nexport * from \"./types.ts\";\nexport * from \"./utils.ts\";\n\nimport type { Page, SearchOptions, SearchResult } from \"./types.ts\";\n// import { searchStrategy, generateEmbeddings } from \"./placeholder.ts\";\n// import { searchStrategy, generateEmbeddings } from \"./jigsawstack-orama.ts\"; // ~550ms query embeddings (API)\n// import { searchStrategy, generateEmbeddings } from \"./openai-orama.ts\"; // ~100-200ms query embeddings (API), ~29ms DB overhead\n// import { searchStrategy, generateEmbeddings } from \"./transformers-cosine.ts\"; // ~10-30ms (browser only), requires --allow-ffi in Deno\nimport { searchStrategy, generateEmbeddings } from \"./openai-cosine.ts\"; // ~100-200ms query embeddings (API), <1ms cosine calc\n\n// Main search function - comment/uncomment imports above to switch strategies\nexport const searchPages = async (\n query: string,\n pages: Page[],\n options: SearchOptions = {}\n): Promise => {\n return searchStrategy.search(query, pages, options);\n};\n\n// Export search strategy for introspection (e.g., /search/test endpoint)\nexport { searchStrategy };\n\n// Embeddings function for recalculate - uses same import as search strategy\nexport const getActiveEmbeddingsFunction = (): ((content: string) => Promise) => {\n return generateEmbeddings;\n};\n\n\n",[],[222,400],"7f271a00-b9ca-11f0-8b68-0224a6c84d84","7f049ff2-b9ca-11f0-98dd-0224a6c84d84",["D",1762293981620],[],[222,412,413,32],"models","all-MiniLM-L6-v2","7f049f52-b9ca-11f0-98dd-0224a6c84d84","7eec2a9e-b9ca-11f0-8b68-0224a6c84d84",51,["D",1762293981385],[254],[222,412,413],"7eec29cc-b9ca-11f0-8b68-0224a6c84d84",50,["D",1762293981234],[253],[222,412],["D",1762207521254],[252,255,256,257,258,259,260,261,262],[222],["D",1762207521817],[],[228],["D",1762293980863],[],[234],["D",1746217442601],{"_94":93,"_101":97},{"_94":93,"_95":-5,"_96":97,"_18":53,"_98":99,"_100":24},{"_13":83,"_31":84,"_41":40,"_85":42,"_86":-5,"_87":-5,"_88":-5,"_89":-5,"_90":-5,"_91":43,"_92":93},"ctx",{"_76":-5,"_121":471,"_472":473,"_474":475,"_476":24,"_477":24,"_478":24,"_479":47,"_480":-5,"_481":-7,"_482":-7,"_483":60,"_484":24,"_485":24,"_486":-5,"_487":24,"_488":24,"_489":24,"_490":-5},"ENV",{"_442":443,"_444":445,"_446":84,"_447":448,"_449":450,"_451":452,"_453":454,"_455":456,"_457":458,"_459":460,"_461":462,"_463":464,"_465":466,"_467":468,"_469":470},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","811aa22498c9b0b5a47e8d778de579cdb990f332","RENDER_GIT_BRANCH","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/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

Diff · v52