Multi-AI · מערכת חכמה

\n \n

שיחה חדשה

\n \n

\n \n \n \n \n \n \n \n \n \n

\n \n

\n \n \n \n

Enter לשליחה · Shift+Enter לשורה חדשה

\n \n

\n\n\n\n\n\n`;\n return new Response(html, {\n headers: {\n \"Content-Type\": \"text/html; charset=utf-8\",\n \"Cache-Control\": \"no-store\",\n },\n });\n}","label",[156],"c0e798c0-bd21-45b1-98ac-d9d4e416bf6b","main.ts",["D",1789579732258],100,"http","// ============================================\n// main.ts - v33 + DeepSeek + 4 Keyless Providers\n// ============================================\n\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { renderPage } from \"./page.ts\";\n\nexport interface AIR { p: string; pk: string; a: string; e?: string; ms?: number; ti?: number; to?: number; }\nexport interface Msg { role: \"user\" | \"assistant\"; content: string; ts?: number; }\nexport interface PL { [key: string]: string; }\n\nconst MODELS: Record = {\n gemini: \"gemini-3.8-flash\",\n groq: \"openai/gpt-oss-120b\",\n deepseek: \"deepseek-chat\",\n cerebras: \"gpt-oss-120b\",\n nvidia: \"meta/llama-3.3-70b-instruct\",\n mistral: \"mistral-small-latest\",\n openrouter: \"nvidia/nemotron-3-super-120b-a12b:free\",\n cohere: \"command-r-plus\",\n zai: \"glm-4-flash\",\n dahl: \"MiniMaxAI/MiniMax-M2.7\",\n llm7: \"deepseek-r1\",\n zen: \"deepseek-v4-flash-free\",\n poll: \"openai\",\n keyless: \"gpt-4o\",\n};\n\nconst NANO_BANANA = \"gemini-2.5-flash-image\";\nconst NANO_BANANA_PRO = \"gemini-3-pro-image\";\nconst NANO_BANANA_2 = \"gemini-3.1-flash-image\";\nconst NANO_BANANA_2_LITE = \"gemini-3.1-flash-lite-image\";\nconst JUDGE_MODEL = \"gemini-3.1-pro-preview\";\nconst TITLE_MODEL = \"gemini-3.5-flash-lite\";\n\nconst CF_MODEL = \"@cf/meta/llama-3.3-70b-instruct-fp8-fast\";\nconst CF_IMAGE_MODEL = \"@cf/black-forest-labs/flux-1-schnell\";\nconst POLLINATIONS_BASE = \"https://gen.pollinations.ai\";\nconst LUMENFALL_BASE = \"https://api.lumenfall.ai/openai/v1/images/generations\";\nconst HUGGINGFACE_BASE = \"https://api-inference.huggingface.co/models\";\n\nconst LUMENFALL_MODELS: Record = {\n draft: \"flux.1-schnell-fp8\",\n hd: \"flux-2-pro\",\n ultra: \"nano-banana-2\",\n};\n\nconst HF_MODELS: Record = {\n draft: \"black-forest-labs/FLUX.1-schnell\",\n hd: \"black-forest-labs/FLUX.1-dev\",\n ultra: \"black-forest-labs/FLUX.1-dev\",\n};\n\nconst P_MS = 12000, J_MS = 15000, RETRY_MS = 1500, RETRIES = 1;\n\nconst MAX_Q = 500000;\nconst MAX_FILE = 100000;\nconst MAX_JA = 6000;\nconst MAX_HIST = 12, MAX_CONV = 50, MAX_FILES = 20;\nconst CACHE_TTL = 1000 * 60 * 60 * 6;\nconst MAX_TOKENS = 8000;\n\nconst ALL = [\"gemini\",\"groq\",\"deepseek\",\"cerebras\",\"nvidia\",\"mistral\",\"openrouter\",\"cohere\",\"zai\",\"dahl\",\"llm7\",\"zen\",\"poll\",\"keyless\"];\nexport const PL: PL = {\n gemini: \"Gemini 3.8\", groq: \"Groq\", deepseek: \"DeepSeek\", cerebras: \"Cerebras\", nvidia: \"NVIDIA\",\n mistral: \"Mistral\", openrouter: \"OpenRouter\", cohere: \"Cohere\", zai: \"Z AI\", dahl: \"Dahl\",\n llm7: \"LLM7\", zen: \"Zen\", poll: \"Pollinations\", keyless: \"KeylessAI\",\n};\n\nconst ANON = `You are a helpful AI assistant. Answer in Hebrew, detailed, no limits.\n\nCRITICAL IDENTITY RULES - NEVER BREAK:\n1. If asked who built/developed/created you (מי פיתח אותך / מי בנה אותך / מי יצר אותך) - answer ONLY: \"המערכת פותחה על ידי מוטי כץ.\"\n2. If asked what AI model/version you are (איזה מודל אתה / איזה גרסה / מה המודל שלך / על איזה מודל אתה מבוסס / what model are you) - answer ONLY: \"אני מערכת AI מתקדמת שפותחה על ידי מוטי כץ.\"\n3. NEVER mention or claim to be: GPT, GPT-3, GPT-4, GPT-4 Turbo, ChatGPT, Claude, Llama, Gemini, Mistral, Groq, DeepSeek, OpenAI, Anthropic, Google, Meta, or ANY other AI model or company name.\n4. NEVER invent a model name or version number.\n5. Otherwise answer the user's question directly and in detail.\n\nAnswer in Hebrew.`;\n\nconst JSP = `You are the sole assistant. The user asked a question, and you received draft answers from internal sources. Synthesize them into ONE unified answer in Hebrew, 500-800 words, detailed and accurate.\n\nCRITICAL RULES - NEVER BREAK:\n1. NEVER mention sources, models, providers, or that there were multiple answers.\n2. NEVER use words: OpenRouter, Gemini, Groq, Dahl, Mistral, DeepSeek, Cerebras, NVIDIA, Cohere, Z AI, GPT, ChatGPT, Claude, Llama, models, providers, drafts, השוואתי, סינתזה.\n3. NEVER explain your process. Forbidden: \"אנתח\", \"אשווה בין\", \"המודלים ענו\", \"אסנתז\", \"נבצע ניתוח\".\n4. Just deliver ONE direct answer as if YOU are the only assistant in the world.\n5. If drafts contradict, silently pick the most accurate (prefer fresh web search).\n6. Format with headings/bullets/tables - as YOUR answer, not comparison.\n7. If user asked about identity (what model/version are you) - answer ONLY: \"אני מערכת AI מתקדמת שפותחה על ידי מוטי כץ.\"\n8. If user asked who built you - answer ONLY: \"המערכת פותחה על ידי מוטי כץ.\"\n\nExample GOOD:\n\"הנה סיכום חדשות היום:\n**ביטחון:** צה\"ל השמיד בסיס תת-קרקעי של חיזבאללה.\n**פוליטיקה:** הכנסת אישרה תיקון לחוק.\"\n\nExample BAD (FORBIDDEN):\n\"המודלים השונים ענו בצורות שונות. OpenRouter היה מפורט, Gemini היה תמציתי.\"`;\n\nconst rateLimitMap = new Map();\nfunction checkRateLimit(cid: string): boolean {\n const now = Date.now();\n const arr = rateLimitMap.get(cid) || [];\n const recent = arr.filter((t) => now - t < 60000);\n if (recent.length >= 30) return false;\n recent.push(now); rateLimitMap.set(cid, recent); return true;\n}\nasync function wt(p: Promise, ms: number, fb: T): Promise {\n let t: any;\n const tp = new Promise((r) => { t = setTimeout(() => r(fb), ms); });\n try { return await Promise.race([p, tp]); } finally { if (t) clearTimeout(t); }\n}\nasync function fr(url: string, o: RequestInit, mr = RETRIES): Promise {\n let last: Response | null = null;\n for (let i = 0; i <= mr; i++) {\n const r = await fetch(url, o);\n if (r.status !== 429 && r.status !== 402) return r;\n last = r;\n if (i < mr) await new Promise((x) => setTimeout(x, RETRY_MS));\n }\n return last as Response;\n}\nfunction tr(s: string, n: number): string {\n return !s ? \"\" : s.length <= n ? s : s.slice(0, n) + \"...\";\n}\nfunction et(s: string): number { return s ? Math.max(1, Math.round(s.length / 4)) : 0; }\nfunction sj(d: unknown): string {\n return JSON.stringify(d).replace(//g, \"\\\\u003e\")\n .replace(/&/g, \"\\\\u0026\").replace(/\\u2028/g, \"\\\\u2028\").replace(/\\u2029/g, \"\\\\u2029\");\n}\nfunction qHash(q: string): string {\n return q.toLowerCase().trim().replace(/\\s+/g, \" \").slice(0, 300);\n}\nfunction isNewsQuery(q: string): boolean {\n const l = q.toLowerCase();\n return l.includes(\"חדשות\") || l.includes(\"היום\") || l.includes(\"news\");\n}\nfunction isMetaQuestion(q: string): boolean {\n const l = q.toLowerCase().trim();\n const patterns = [\n \"איזה מודל\", \"איזה גרסה\", \"מה המודל\", \"מה הגרסה\", \"איזה מודל אתה\",\n \"מי פיתח\", \"מי בנה\", \"מי יצר\", \"מי אתה\", \"מה אתה\",\n \"על איזה מודל\", \"על איזה גרסה\", \"what model\", \"which version\",\n \"what version\", \"who made you\", \"who built you\", \"who created you\",\n ];\n return patterns.some((k) => l.includes(k));\n}\nfunction validateAnswer(answer: string): boolean {\n if (!answer || answer.length < 15) return false;\n const trimmed = answer.trim();\n if (/^Source:\\s/m.test(trimmed)) return false;\n if (/^URL:\\s/m.test(trimmed)) return false;\n if (trimmed.startsWith(\"Source:\") || trimmed.startsWith(\"--- \")) return false;\n if (trimmed.includes(\"support.apple.com\") || trimmed.includes(\"apple.com/he-il\")) return false;\n if (trimmed.includes(\"iPhone 1\") || trimmed.includes(\"iOS 26\")) return false;\n const forbidden = [\"OpenRouter said\", \"Gemini said\", \"Groq said\", \"Dahl said\", \"Mistral said\", \"Cerebras said\", \"DeepSeek said\"];\n if (forbidden.some((f) => trimmed.includes(f))) return false;\n const urlCount = (trimmed.match(/https?:\\/\\//g) || []).length;\n if (urlCount > 3 && urlCount * 100 > trimmed.length) return false;\n return true;\n}\n\nconst translationCache = new Map();\nconst TRANSLATION_CACHE_MAX = 200;\nfunction cacheTranslation(src: string, dst: string) {\n if (translationCache.size >= TRANSLATION_CACHE_MAX) {\n const firstKey = translationCache.keys().next().value;\n if (firstKey) translationCache.delete(firstKey);\n }\n translationCache.set(src, dst);\n}\n\nasync function dbInit() {\n try {\n await sqlite.execute(\"CREATE TABLE IF NOT EXISTS conversations (id INTEGER PRIMARY KEY AUTOINCREMENT, conversation_id TEXT NOT NULL, role TEXT NOT NULL, content TEXT NOT NULL, created_at INTEGER NOT NULL)\");\n await sqlite.execute(\"CREATE INDEX IF NOT EXISTS idx_conv_id ON conversations(conversation_id)\");\n await sqlite.execute(\"CREATE TABLE IF NOT EXISTS conv_meta (conversation_id TEXT PRIMARY KEY, title TEXT NOT NULL, created_at INTEGER NOT NULL, updated_at INTEGER NOT NULL)\");\n await sqlite.execute(\"CREATE TABLE IF NOT EXISTS raw_answers (id INTEGER PRIMARY KEY AUTOINCREMENT, conversation_id TEXT NOT NULL, question TEXT NOT NULL, provider TEXT NOT NULL, provider_label TEXT NOT NULL, answer TEXT NOT NULL, created_at INTEGER NOT NULL)\");\n await sqlite.execute(\"CREATE INDEX IF NOT EXISTS idx_raw_conv ON raw_answers(conversation_id)\");\n await sqlite.execute(\"CREATE TABLE IF NOT EXISTS cache (qhash TEXT PRIMARY KEY, question TEXT NOT NULL, answer TEXT NOT NULL, raw_json TEXT NOT NULL, created_at INTEGER NOT NULL)\");\n await sqlite.execute(\"CREATE TABLE IF NOT EXISTS provider_health (provider TEXT PRIMARY KEY, last_success INTEGER, last_failure INTEGER, failure_count INTEGER DEFAULT 0)\");\n await sqlite.execute(\"CREATE TABLE IF NOT EXISTS title_cache (qhash TEXT PRIMARY KEY, title TEXT NOT NULL, created_at INTEGER NOT NULL)\");\n try { await sqlite.execute(\"ALTER TABLE conv_meta ADD COLUMN pinned INTEGER DEFAULT 0\"); } catch {}\n } catch {}\n}\nasync function dbHist(cid: string): Promise {\n if (!cid) return [];\n try {\n const r = await sqlite.execute({ sql: \"SELECT role, content FROM conversations WHERE conversation_id =? ORDER BY created_at DESC LIMIT?\", args: [cid, MAX_HIST] });\n return (r.rows as any[] || []).map((row) => ({ role: String(row.role) === \"user\" ? \"user\" : \"assistant\", content: String(row.content || \"\") } as Msg)).filter((m) => m.content).reverse();\n } catch { return []; }\n}\nasync function dbSave(cid: string, role: string, content: string) {\n if (!cid || !content) return;\n try { await sqlite.execute({ sql: \"INSERT INTO conversations (conversation_id, role, content, created_at) VALUES (?,?,?,?)\", args: [cid, role, content, Date.now()] }); } catch {}\n}\nasync function dbSaveRaw(cid: string, question: string, raws: AIR[]) {\n if (!cid || !raws.length) return;\n const now = Date.now();\n try {\n for (const r of raws) {\n if (!r.a) continue;\n await sqlite.execute({ sql: \"INSERT INTO raw_answers (conversation_id, question, provider, provider_label, answer, created_at) VALUES (?,?,?,?,?,?)\", args: [cid, question, r.pk, r.p, r.a, now] });\n }\n } catch {}\n}\nasync function dbGetRaw(cid: string) {\n try {\n const r = await sqlite.execute({ sql: \"SELECT question, provider, provider_label, answer, created_at FROM raw_answers WHERE conversation_id =? ORDER BY created_at ASC\", args: [cid] });\n return (r.rows as any[] || []).map((row) => ({ question: String(row.question || \"\"), provider: String(row.provider || \"\"), label: String(row.provider_label || \"\"), answer: String(row.answer || \"\"), ts: Number(row.created_at) || 0 }));\n } catch { return []; }\n}\nasync function getCache(qh: string) {\n try {\n const r = await sqlite.execute({ sql: \"SELECT answer, raw_json, created_at FROM cache WHERE qhash=?\", args: [qh] });\n const row = (r.rows as any[])?.[0];\n if (!row) return null;\n if (Date.now() - Number(row.created_at) > CACHE_TTL) return null;\n return { answer: String(row.answer), raw: JSON.parse(String(row.raw_json || \"[]\")) };\n } catch { return null; }\n}\nasync function setCache(qh: string, q: string, answer: string, raw: AIR[]) {\n try { await sqlite.execute({ sql: \"INSERT OR REPLACE INTO cache (qhash, question, answer, raw_json, created_at) VALUES (?,?,?,?,?)\", args: [qh, q, answer, JSON.stringify(raw), Date.now()] }); } catch {}\n}\nasync function dbMeta(cid: string, title: string) {\n if (!cid) return;\n try {\n const ex = await sqlite.execute({ sql: \"SELECT conversation_id FROM conv_meta WHERE conversation_id =?\", args: [cid] });\n const now = Date.now();\n const safeTitle = tr(title, 60) || \"שיחה חדשה\";\n if ((ex.rows as any[]).length === 0) {\n await sqlite.execute({ sql: \"INSERT INTO conv_meta (conversation_id, title, created_at, updated_at, pinned) VALUES (?,?,?,?,?)\", args: [cid, safeTitle, now, now, 0] });\n } else {\n await sqlite.execute({ sql: \"UPDATE conv_meta SET updated_at =?, title = COALESCE(title,?) WHERE conversation_id =?\", args: [now, safeTitle, cid] });\n }\n } catch {}\n}\nasync function dbList() {\n try {\n const r = await sqlite.execute({ sql: \"SELECT conversation_id, title, updated_at, pinned FROM conv_meta ORDER BY pinned DESC, updated_at DESC LIMIT?\", args: [MAX_CONV] });\n return (r.rows as any[] || []).map((row) => ({ id: String(row.conversation_id || \"\"), title: String(row.title || \"ללא כותרת\"), updated_at: Number(row.updated_at) || 0, pinned: Number(row.pinned) || 0 }));\n } catch { return []; }\n}\nasync function dbFull(cid: string): Promise {\n if (!cid) return [];\n try {\n const r = await sqlite.execute({ sql: \"SELECT role, content, created_at FROM conversations WHERE conversation_id =? ORDER BY created_at ASC\", args: [cid] });\n return (r.rows as any[] || []).map((row) => ({ role: String(row.role) === \"user\" ? \"user\" : \"assistant\", content: String(row.content || \"\"), ts: Number(row.created_at) || 0 } as Msg)).filter((m) => m.content);\n } catch { return []; }\n}\nasync function dbDel(cid: string) {\n if (!cid) return false;\n try {\n await sqlite.execute({ sql: \"DELETE FROM conversations WHERE conversation_id =?\", args: [cid] });\n await sqlite.execute({ sql: \"DELETE FROM conv_meta WHERE conversation_id =?\", args: [cid] });\n await sqlite.execute({ sql: \"DELETE FROM raw_answers WHERE conversation_id =?\", args: [cid] });\n return true;\n } catch { return false; }\n}\nasync function dbRen(cid: string, title: string) {\n if (!cid || !title) return false;\n try { await sqlite.execute({ sql: \"UPDATE conv_meta SET title =?, updated_at =? WHERE conversation_id =?\", args: [tr(title, 80), Date.now(), cid] }); return true; } catch { return false; }\n}\nasync function dbHealth(provider: string, success: boolean) {\n try {\n const now = Date.now();\n const ex = await sqlite.execute({ sql: \"SELECT provider FROM provider_health WHERE provider =?\", args: [provider] });\n if ((ex.rows as any[]).length === 0) {\n await sqlite.execute({ sql: \"INSERT INTO provider_health (provider, last_success, last_failure, failure_count) VALUES (?,?,?,?)\", args: [provider, success ? now : 0, success ? 0 : now, success ? 0 : 1] });\n } else if (success) {\n await sqlite.execute({ sql: \"UPDATE provider_health SET last_success =?, failure_count = 0 WHERE provider =?\", args: [now, provider] });\n } else {\n await sqlite.execute({ sql: \"UPDATE provider_health SET last_failure =?, failure_count = failure_count + 1 WHERE provider =?\", args: [now, provider] });\n }\n } catch {}\n}\nasync function dbGetHealth() {\n try {\n const r = await sqlite.execute({ sql: \"SELECT provider, last_success, last_failure, failure_count FROM provider_health\" });\n return (r.rows as any[] || []).map((row) => ({ provider: String(row.provider), last_success: Number(row.last_success) || 0, last_failure: Number(row.last_failure) || 0, failure_count: Number(row.failure_count) || 0 }));\n } catch { return []; }\n}\nasync function getCachedTitle(qh: string): Promise {\n try {\n const r = await sqlite.execute({ sql: \"SELECT title, created_at FROM title_cache WHERE qhash=?\", args: [qh] });\n const row = (r.rows as any[])?.[0];\n if (!row) return null;\n if (Date.now() - Number(row.created_at) > CACHE_TTL) return null;\n return String(row.title);\n } catch { return null; }\n}\nasync function setCachedTitle(qh: string, title: string) {\n try { await sqlite.execute({ sql: \"INSERT OR REPLACE INTO title_cache (qhash, title, created_at) VALUES (?,?,?)\", args: [qh, title, Date.now()] }); } catch {}\n}\n\nasync function generateTitle(q: string, k: string): Promise {\n if (!k) return tr(q, 40);\n const qh = \"title:\" + qHash(q);\n const cached = await getCachedTitle(qh);\n if (cached) return cached;\n try {\n const url = \"https://generativelanguage.googleapis.com/v1beta/models/\" + TITLE_MODEL + \":generateContent?key=\" + encodeURIComponent(k);\n const r = await fr(url, {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n contents: [{ parts: [{ text: 'כותרת 3-4 מילים בעברית (בלי מירכאות): \"' + tr(q, 200) + '\"' }] }],\n generationConfig: { temperature: 0.3, maxOutputTokens: 20 },\n }),\n }, 0);\n if (!r.ok) return tr(q, 40);\n const d = await r.json();\n const title = d?.candidates?.[0]?.content?.parts?.[0]?.text?.trim().replace(/[\"']/g, \"\").slice(0, 40) || tr(q, 40);\n if (title && title !== tr(q, 40)) await setCachedTitle(qh, title);\n return title;\n } catch { return tr(q, 40); }\n}\nfunction ctx(hist: Msg[], q: string): string {\n if (hist.length === 0) return q;\n return tr(\"Previous:\\n\" + hist.map((m) => (m.role === \"user\" ? \"User: \" : \"Assistant: \") + m.content).join(\"\\n\") + \"\\n\\nQuestion: \" + q, 15000);\n}\nasync function webSearch(q: string, k: string): Promise {\n if (!k || !q) return \"\";\n try {\n const today = new Date().toISOString().split(\"T\")[0];\n const searchQ = isNewsQuery(q) ? q + \" \" + today + \" Israel\" : q;\n const r = await fetch(\"https://api.tavily.com/search\", {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n api_key: k, query: searchQ, max_results: 8, search_depth: \"advanced\",\n time_range: isNewsQuery(q) ? \"day\" : undefined, include_answer: true,\n }),\n });\n if (!r.ok) return \"\";\n const d = await r.json();\n return (d.results as any[] || []).map((x) => \"Source: \" + x.title + \"\\n\" + x.content + \"\\nURL: \" + x.url).join(\"\\n\\n\") + (d.answer ? \"\\n\\nSummary: \" + d.answer : \"\");\n } catch { return \"\"; }\n}\nasync function translateToEnglish(p: string, k: string): Promise {\n if (!k || !p) return p;\n if (/^[a-zA-Z0-9,.'\"\\-\\s]+$/.test(p) && p.length < 100) return p;\n const cached = translationCache.get(p);\n if (cached) return cached;\n try {\n const url = \"https://generativelanguage.googleapis.com/v1beta/models/\" + TITLE_MODEL + \":generateContent?key=\" + encodeURIComponent(k);\n const r = await fr(url, {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n contents: [{ parts: [{\n text: `You are a professional Hebrew-to-English translator for AI image prompts.\n\nTASK: Translate the Hebrew text below into natural English.\n\nRULES:\n1. Output ONLY the English translation.\n2. Preserve ALL details - never summarize, truncate, or omit words.\n3. Keep the complete sentence structure.\n4. Do NOT add quotes, prefixes, or explanations.\n5. If some words are already English, keep them as-is.\n6. Add commas between descriptive clauses.\n\nHebrew text: \"${p}\"\n\nEnglish translation:`\n }] }],\n generationConfig: { temperature: 0.1, maxOutputTokens: 500 },\n }),\n }, 0);\n if (!r.ok) return p;\n const d = await r.json();\n const translated = d?.candidates?.[0]?.content?.parts?.[0]?.text?.trim().replace(/^[\"']|[\"']$/g, \"\") || p;\n if (translated.length >= p.length * 0.6) {\n const result = translated.slice(0, 500);\n cacheTranslation(p, result);\n return result;\n }\n const r2 = await fr(url, {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n contents: [{ parts: [{ text: `Translate to English, preserve ALL details. Hebrew: ${p}\\nEnglish:` }] }],\n generationConfig: { temperature: 0.0, maxOutputTokens: 500 },\n }),\n }, 0);\n if (r2.ok) {\n const d2 = await r2.json();\n const t2 = d2?.candidates?.[0]?.content?.parts?.[0]?.text?.trim().replace(/^[\"']|[\"']$/g, \"\") || \"\";\n if (t2.length > translated.length) {\n const result = t2.slice(0, 500);\n cacheTranslation(p, result);\n return result;\n }\n }\n const final = translated.slice(0, 500);\n cacheTranslation(p, final);\n return final;\n } catch { return p; }\n}\nasync function translateToEnglishSafe(p: string, k: string): Promise {\n const primary = await translateToEnglish(p, k);\n if (p.length > 20 && primary.length < p.length * 0.6) {\n try {\n const chunks = p.split(/[\\s,،;.\\n]+/).map((s) => s.trim()).filter((s) => s.length > 1);\n if (chunks.length > 1) {\n const parts: string[] = [];\n for (const c of chunks) {\n const t = await translateToEnglish(c, k);\n parts.push(t);\n }\n const joined = parts.join(\", \");\n if (joined.length > primary.length * 1.2) return joined.slice(0, 500);\n }\n } catch {}\n }\n return primary;\n}\n\nlet pollinationsCatalog: any[] | null = null;\nconst aliasToCanonical = new Map();\nconst canonicalToMeta = new Map();\nasync function fetchPollinationsCatalog(): Promise {\n if (pollinationsCatalog) return pollinationsCatalog;\n try {\n const r = await fetch(POLLINATIONS_BASE + \"/models\");\n if (!r.ok) return [];\n const data = await r.json();\n const list = Array.isArray(data) ? data : (data?.models || data?.data || []);\n pollinationsCatalog = list;\n aliasToCanonical.clear(); canonicalToMeta.clear();\n for (const m of list) {\n const canonical = m.id || m.name || \"\";\n if (!canonical) continue;\n const meta = { id: canonical, publisher: m.publisher || m.brand || canonical.split(\"/\")[0], brand_url: m.brand_url || \"\", aliases: m.aliases || [], ...m };\n canonicalToMeta.set(canonical, meta);\n aliasToCanonical.set(canonical, canonical);\n for (const alias of (m.aliases || [])) aliasToCanonical.set(alias, canonical);\n const short = canonical.split(\"/\").pop()!;\n if (!aliasToCanonical.has(short)) aliasToCanonical.set(short, canonical);\n }\n return list;\n } catch { return []; }\n}\nfunction resolveModelId(input: string): string {\n if (aliasToCanonical.has(input)) return aliasToCanonical.get(input)!;\n const low = input.toLowerCase();\n for (const [k, v] of aliasToCanonical.entries()) if (k.toLowerCase() === low) return v;\n return input;\n}\nfunction getModelMeta(id: string) {\n const canonical = resolveModelId(id);\n return canonicalToMeta.get(canonical) || { id: canonical, publisher: canonical.split(\"/\")[0], brand_url: \"\" };\n}\n\nasync function genImage(\n promptEn: string,\n acc: string,\n tok: string,\n style: string = \"photorealistic\",\n quality: string = \"hd\",\n geminiKey: string = \"\",\n pollinationsKey: string = \"\",\n lumenfallKey: string = \"\",\n huggingfaceKey: string = \"\",\n): Promise {\n const pk = \"image\", p = \"🎨 ULTRA\";\n const seed = Math.floor(Math.random() * 999999) + 1;\n let width = 1280, height = 1280, steps = 20;\n if (quality === \"ultra\") { width = 1536; height = 1536; steps = 30; }\n else if (quality === \"draft\") { width = 768; height = 768; steps = 12; }\n\n let suffix = \", ultra detailed, 8k, masterpiece, photorealistic, cinematic lighting, sharp focus, best quality\";\n if (style === \"anime\") suffix = \", anime style, vibrant, 8k\";\n else if (style === \"3d\") suffix = \", 3d render, octane, 8k\";\n else if (style === \"fast\") suffix = \", quick sketch, minimal\";\n const enhanced = promptEn + suffix;\n const enc = encodeURIComponent(enhanced);\n\n if (geminiKey && quality === \"ultra\") {\n try {\n const url = \"https://generativelanguage.googleapis.com/v1beta/models/\" + NANO_BANANA_PRO + \":generateContent?key=\" + encodeURIComponent(geminiKey);\n const r = await fr(url, {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ contents: [{ parts: [{ text: enhanced }] }], generationConfig: { responseModalities: [\"IMAGE\", \"TEXT\"] } }),\n }, 0);\n if (r.ok) {\n const d = await r.json();\n const parts = d?.candidates?.[0]?.content?.parts || [];\n for (const part of parts) {\n if (part.inlineData && part.inlineData.data) {\n console.log(\"[IMG] Layer 1 (Nano Banana Pro) OK\");\n return { p: \"🍌 ננו-בננה PRO\", pk, a: \"![תמונה](data:\" + (part.inlineData.mimeType || \"image/png\") + \";base64,\" + part.inlineData.data + \")\", ms: 0 };\n }\n }\n }\n console.log(\"[IMG] Layer 1 failed status:\", r.status);\n } catch (e) { console.log(\"[IMG] Layer 1 error:\", String(e)); }\n }\n\n if (geminiKey && quality !== \"draft\") {\n try {\n const url = \"https://generativelanguage.googleapis.com/v1beta/models/\" + NANO_BANANA_2 + \":generateContent?key=\" + encodeURIComponent(geminiKey);\n const r = await fr(url, {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ contents: [{ parts: [{ text: enhanced }] }], generationConfig: { responseModalities: [\"IMAGE\", \"TEXT\"] } }),\n }, 0);\n if (r.ok) {\n const d = await r.json();\n const parts = d?.candidates?.[0]?.content?.parts || [];\n for (const part of parts) {\n if (part.inlineData && part.inlineData.data) {\n console.log(\"[IMG] Layer 2 (Nano Banana 2) OK\");\n return { p: \"🍌 ננו-בננה 2\", pk, a: \"![תמונה](data:\" + (part.inlineData.mimeType || \"image/png\") + \";base64,\" + part.inlineData.data + \")\", ms: 0 };\n }\n }\n }\n console.log(\"[IMG] Layer 2 failed status:\", r.status);\n } catch (e) { console.log(\"[IMG] Layer 2 error:\", String(e)); }\n }\n\n if (geminiKey && quality === \"draft\") {\n try {\n const url = \"https://generativelanguage.googleapis.com/v1beta/models/\" + NANO_BANANA_2_LITE + \":generateContent?key=\" + encodeURIComponent(geminiKey);\n const r = await fr(url, {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ contents: [{ parts: [{ text: enhanced }] }], generationConfig: { responseModalities: [\"IMAGE\", \"TEXT\"] } }),\n }, 0);\n if (r.ok) {\n const d = await r.json();\n const parts = d?.candidates?.[0]?.content?.parts || [];\n for (const part of parts) {\n if (part.inlineData && part.inlineData.data) {\n console.log(\"[IMG] Layer 3 (Nano Banana 2 Lite) OK\");\n return { p: \"🍌 ננו-בננה Lite\", pk, a: \"![תמונה](data:\" + (part.inlineData.mimeType || \"image/png\") + \";base64,\" + part.inlineData.data + \")\", ms: 0 };\n }\n }\n }\n console.log(\"[IMG] Layer 3 failed status:\", r.status);\n } catch (e) { console.log(\"[IMG] Layer 3 error:\", String(e)); }\n }\n\n if (geminiKey) {\n try {\n const url = \"https://generativelanguage.googleapis.com/v1beta/models/\" + NANO_BANANA + \":generateContent?key=\" + encodeURIComponent(geminiKey);\n const r = await fr(url, {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ contents: [{ parts: [{ text: enhanced }] }], generationConfig: { responseModalities: [\"IMAGE\", \"TEXT\"] } }),\n }, 0);\n if (r.ok) {\n const d = await r.json();\n const parts = d?.candidates?.[0]?.content?.parts || [];\n for (const part of parts) {\n if (part.inlineData && part.inlineData.data) {\n console.log(\"[IMG] Layer 4 (Nano Banana Classic) OK\");\n return { p: \"🍌 ננו-בננה\", pk, a: \"![תמונה](data:\" + (part.inlineData.mimeType || \"image/png\") + \";base64,\" + part.inlineData.data + \")\", ms: 0 };\n }\n }\n }\n console.log(\"[IMG] Layer 4 failed status:\", r.status);\n } catch (e) { console.log(\"[IMG] Layer 4 error:\", String(e)); }\n }\n\n if (pollinationsKey) {\n try {\n await fetchPollinationsCatalog();\n let desiredModel = \"black-forest-labs/flux.2-pro\";\n if (quality === \"ultra\") desiredModel = \"google/gemini-3-pro-image\";\n else if (quality === \"draft\") desiredModel = \"black-forest-labs/flux.1-schnell\";\n const canonical = resolveModelId(desiredModel);\n const meta = getModelMeta(canonical);\n const r = await fetch(POLLINATIONS_BASE + \"/v1/images/generations\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\", Authorization: \"Bearer \" + pollinationsKey },\n body: JSON.stringify({ model: canonical, prompt: enhanced, width, height, n: 1, response_format: \"url\" }),\n });\n if (r.ok) {\n const d = await r.json();\n const imgUrl = d?.data?.[0]?.url || (d?.data?.[0]?.b64_json ? \"data:image/png;base64,\" + d.data[0].b64_json : \"\");\n if (imgUrl) {\n console.log(\"[IMG] Layer 5 (Pollinations API) OK\");\n return { p: \"🎨 \" + meta.publisher, pk, a: \"![תמונה](\" + imgUrl + \")\", ms: 0 };\n }\n }\n console.log(\"[IMG] Layer 5 failed status:\", r.status);\n } catch (e) { console.log(\"[IMG] Layer 5 error:\", String(e)); }\n }\n\n if (lumenfallKey) {\n try {\n const lumenModel = LUMENFALL_MODELS[quality] || LUMENFALL_MODELS.hd;\n let size = \"1024x1024\";\n if (quality === \"draft\") size = \"512x512\";\n else if (quality === \"ultra\") size = \"1536x1536\";\n const r = await fr(LUMENFALL_BASE, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\", Authorization: \"Bearer \" + lumenfallKey },\n body: JSON.stringify({ model: lumenModel, prompt: enhanced, size, n: 1, response_format: \"url\" }),\n }, 0);\n if (r.ok) {\n const d = await r.json();\n const imgUrl = d?.data?.[0]?.url || \"\";\n if (imgUrl) {\n console.log(\"[IMG] Layer 6 (Lumenfall) OK\");\n return { p: \"⚡ Lumenfall \" + lumenModel, pk, a: \"![תמונה](\" + imgUrl + \")\", ms: 0 };\n }\n }\n console.log(\"[IMG] Layer 6 failed status:\", r.status);\n } catch (e) { console.log(\"[IMG] Layer 6 error:\", String(e)); }\n }\n\n if (huggingfaceKey) {\n try {\n const hfModel = HF_MODELS[quality] || HF_MODELS.hd;\n const url = HUGGINGFACE_BASE + \"/\" + hfModel;\n const r = await fr(url, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\", Authorization: \"Bearer \" + huggingfaceKey },\n body: JSON.stringify({ inputs: enhanced, parameters: { width, height } }),\n }, 0);\n if (r.ok) {\n const buf = await r.arrayBuffer();\n if (buf.byteLength > 5000) {\n const bytes = new Uint8Array(buf);\n let binary = \"\";\n const chunk = 8192;\n for (let i = 0; i < bytes.length; i += chunk) {\n binary += String.fromCharCode.apply(null, Array.from(bytes.subarray(i, i + chunk)) as any);\n }\n console.log(\"[IMG] Layer 7 (HuggingFace) OK\");\n return { p: \"🤗 HuggingFace \" + hfModel.split(\"/\").pop(), pk, a: \"![תמונה](data:image/png;base64,\" + btoa(binary) + \")\", ms: 0 };\n }\n }\n console.log(\"[IMG] Layer 7 failed status:\", r.status);\n } catch (e) { console.log(\"[IMG] Layer 7 error:\", String(e)); }\n }\n\n if (acc && tok) {\n try {\n const url = \"https://api.cloudflare.com/client/v4/accounts/\" + encodeURIComponent(acc) + \"/ai/run/\" + CF_IMAGE_MODEL;\n const r = await fr(url, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\", Authorization: \"Bearer \" + tok },\n body: JSON.stringify({ prompt: enhanced, num_steps: steps, guidance: 7.5 }),\n }, 0);\n if (r.ok) {\n const buf = await r.arrayBuffer();\n if (buf.byteLength > 5000) {\n const bytes = new Uint8Array(buf);\n let binary = \"\";\n const chunk = 8192;\n for (let i = 0; i < bytes.length; i += chunk) {\n binary += String.fromCharCode.apply(null, Array.from(bytes.subarray(i, i + chunk)) as any);\n }\n console.log(\"[IMG] Layer 8 (Cloudflare) OK\");\n return { p: \"☁️ Cloudflare FLUX\", pk, a: \"![תמונה](data:image/png;base64,\" + btoa(binary) + \")\", ms: 0 };\n }\n }\n console.log(\"[IMG] Layer 8 failed status:\", r.status);\n } catch (e) { console.log(\"[IMG] Layer 8 error:\", String(e)); }\n }\n\n console.log(\"[IMG] Layer 9 (URL fallback) - all layers failed\");\n const proUrl = \"https://image.pollinations.ai/prompt/\" + enc + \"?width=\" + width + \"&height=\" + height + \"&model=flux-pro&enhance=true&nologo=true&seed=\" + seed;\n return { p: \"🔗 Pollinations URL\", pk, a: \"![תמונה](\" + proUrl + \")\", e: \"רק fallback עבד - בדוק מפתחות\" };\n}\n\nasync function askGemini(prompt: string, k: string, model: string = MODELS.gemini): Promise {\n const pk = \"gemini\", p = PL[pk];\n if (!k) return { p, pk, a: \"\", e: \"לא מוגדר\" };\n const s = Date.now();\n return wt((async () => {\n try {\n const url = \"https://generativelanguage.googleapis.com/v1beta/models/\" + model + \":generateContent?key=\" + encodeURIComponent(k);\n const r = await fr(url, {\n method: \"POST\", headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n systemInstruction: { parts: [{ text: ANON }] },\n contents: [{ parts: [{ text: prompt }] }],\n generationConfig: { temperature: 0.7, maxOutputTokens: MAX_TOKENS },\n }),\n });\n if (!r.ok) { await dbHealth(pk, false); return { p, pk, a: \"\", e: \"שגיאה \" + r.status, ms: Date.now() - s }; }\n const d = await r.json();\n const a = d?.candidates?.[0]?.content?.parts?.[0]?.text?.trim() || \"\";\n if (!a) { await dbHealth(pk, false); return { p, pk, a: \"\", e: \"ריק\", ms: Date.now() - s }; }\n await dbHealth(pk, true);\n return { p, pk, a, ms: Date.now() - s, ti: et(prompt), to: et(a) };\n } catch (e) {\n await dbHealth(pk, false);\n return { p, pk, a: \"\", e: String(e).slice(0, 120), ms: Date.now() - s };\n }\n })(), P_MS, { p, pk, a: \"\", e: \"פסק זמן\", ms: 0 });\n}\nasync function askCompat(prompt: string, k: string, pk: string, endpoint: string, model: string): Promise {\n const p = PL[pk];\n if (!k) return { p, pk, a: \"\", e: \"לא מוגדר\" };\n const s = Date.now();\n return wt((async () => {\n try {\n const h: any = { \"Content-Type\": \"application/json\", Authorization: \"Bearer \" + k };\n if (pk === \"openrouter\") { h[\"HTTP-Referer\"] = \"https://multi-ai-chat.com\"; h[\"X-Title\"] = \"Multi-AI\"; }\n const r = await fr(endpoint, {\n method: \"POST\", headers: h,\n body: JSON.stringify({\n model, messages: [{ role: \"system\", content: ANON }, { role: \"user\", content: prompt }],\n temperature: 0.7, max_tokens: MAX_TOKENS,\n }),\n });\n if (!r.ok) { await dbHealth(pk, false); return { p, pk, a: \"\", e: \"שגיאה \" + r.status, ms: Date.now() - s }; }\n const d = await r.json();\n const a = d?.choices?.[0]?.message?.content?.trim() || \"\";\n if (!a) { await dbHealth(pk, false); return { p, pk, a: \"\", e: \"ריק\", ms: Date.now() - s }; }\n await dbHealth(pk, true);\n return { p, pk, a, ms: Date.now() - s, ti: et(prompt), to: et(a) };\n } catch (e) {\n await dbHealth(pk, false);\n return { p, pk, a: \"\", e: String(e).slice(0, 120), ms: Date.now() - s };\n }\n })(), P_MS, { p, pk, a: \"\", e: \"פסק זמן\", ms: 0 });\n}\nlet groqIdx = 0;\nasync function askGroqDual(prompt: string, k1: string, k2: string): Promise {\n const pk = \"groq\", p = PL[pk];\n const keys = [k1, k2].filter(Boolean);\n if (keys.length === 0) return { p, pk, a: \"\", e: \"לא מוגדר\" };\n const eps = \"https://api.groq.com/openai/v1/chat/completions\";\n const errors: string[] = [];\n for (let attempt = 0; attempt < keys.length; attempt++) {\n const idx = (groqIdx + attempt) % keys.length;\n const r = await askCompat(prompt, keys[idx], pk, eps, MODELS[pk]);\n if (r.a && !r.e) {\n groqIdx = (idx + 1) % keys.length;\n if (idx !== 0) r.p = PL[pk] + \" (key\" + (idx + 1) + \")\";\n return r;\n }\n errors.push(\"key\" + (idx + 1) + \": \" + (r.e || \"unknown\"));\n if (!(r.e && (r.e.includes(\"429\") || r.e.includes(\"402\")))) return r;\n }\n groqIdx = (groqIdx + 1) % keys.length;\n return { p, pk, a: \"\", e: \"כל המפתחות נכשלו → \" + errors.join(\" | \") };\n}\nasync function askProvider(prompt: string, keys: any, pk: string): Promise {\n if (pk === \"gemini\") return askGemini(prompt, keys.gemini);\n if (pk === \"groq\") return askGroqDual(prompt, keys.groq, keys.groq2);\n const eps: Record = {\n deepseek: \"https://api.deepseek.com/v1/chat/completions\",\n cerebras: \"https://api.cerebras.ai/v1/chat/completions\",\n nvidia: \"https://integrate.api.nvidia.com/v1/chat/completions\",\n mistral: \"https://api.mistral.ai/v1/chat/completions\",\n openrouter: \"https://openrouter.ai/api/v1/chat/completions\",\n cohere: \"https://api.cohere.ai/compatibility/v1/chat/completions\",\n zai: \"https://open.bigmodel.cn/api/paas/v4/chat/completions\",\n dahl: \"https://inference.dahl.global/v1/chat/completions\",\n llm7: \"https://api.llm7.io/v1/chat/completions\",\n zen: \"https://opencode.ai/zen/v1/chat/completions\",\n poll: \"https://text.pollinations.ai/openai/chat/completions\",\n keyless: \"https://keylessai.thryx.workers.dev/v1/chat/completions\",\n };\n return askCompat(prompt, keys[pk], pk, eps[pk], MODELS[pk]);\n}\nasync function cfCall(messages: any[], acc: string, tok: string): Promise {\n try {\n const url = \"https://api.cloudflare.com/client/v4/accounts/\" + encodeURIComponent(acc) + \"/ai/run/\" + CF_MODEL;\n const r = await fr(url, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\", Authorization: \"Bearer \" + tok },\n body: JSON.stringify({ messages, max_tokens: MAX_TOKENS }),\n });\n if (!r.ok) return \"\";\n const d = await r.json();\n return String(d?.result?.response || \"\").trim();\n } catch { return \"\"; }\n}\n\nasync function runJudge(jp: string, acc: string, tok: string, geminiKey: string): Promise {\n if (geminiKey) {\n const g = await askGemini(jp, geminiKey, JUDGE_MODEL);\n if (g.a && !g.e) return g.a;\n }\n const cf = await wt(cfCall([{ role: \"system\", content: JSP }, { role: \"user\", content: jp }], acc, tok), J_MS, \"\");\n if (cf) return cf;\n if (geminiKey) {\n const g2 = await askGemini(jp, geminiKey, MODELS.gemini);\n if (g2.a) return g2.a;\n }\n return \"\";\n}\nfunction buildJP(q: string, answers: AIR[], hist: Msg[], searchCtx: string): string {\n const valid = answers.filter((a) => a.a && a.a.length > 20 && !a.a.includes(\"המערכת פותחה על ידי מוטי כץ\"));\n const fb = valid.length === 0 ? answers.filter((a) => a.a && a.a.length > 0) : valid;\n const txt = fb.map((a) => \"--- DRAFT ---\\n\" + tr(a.a, MAX_JA)).join(\"\\n\\n\");\n const histTxt = hist.length\n ? \"History:\\n\" + hist.map((m) => (m.role === \"user\" ? \"User: \" : \"Assistant: \") + tr(m.content, 500)).join(\"\\n\") + \"\\n\\n\"\n : \"\";\n const searchBlock = searchCtx\n ? \"\\n\\nWEB SEARCH (FRESH \" + new Date().toLocaleDateString(\"he-IL\") + \"):\\n\" + tr(searchCtx, 6000) + \"\\n\"\n : \"\";\n return JSP + \"\\n\\n\" + histTxt + \"User Question: \" + q + \"\\n\" + searchBlock + \"\\nDrafts:\\n\" + txt + \"\\n\\nFinal answer:\";\n}\nfunction parseJP(out: string) {\n const s = String(out || \"\");\n const a = s.match(/ANSWER:\\s*([\\s\\S]*)/i);\n return { answer: a ? a[1].trim() : s.trim() };\n}\n\nexport default async function (req: Request): Promise {\n const url = new URL(req.url);\n const GEMINI_API_KEY = Deno.env.get(\"GEMINI_API_KEY\") || \"\";\n const GROQ_API_KEY = Deno.env.get(\"GROQ_API_KEY\") || \"\";\n const GROQ_API_KEY_2 = Deno.env.get(\"GROQ_API_KEY_2\") || \"\";\n const DEEPSEEK_API_KEY = Deno.env.get(\"DEEPSEEK_API_KEY\") || \"\";\n const CEREBRAS_API_KEY = Deno.env.get(\"CEREBRAS_API_KEY\") || \"\";\n const NVIDIA_API_KEY = Deno.env.get(\"NVIDIA_API_KEY\") || \"\";\n const MISTRAL_API_KEY = Deno.env.get(\"MISTRAL_API_KEY\") || \"\";\n const OPENROUTER_API_KEY = Deno.env.get(\"OPENROUTER_API_KEY\") || \"\";\n const COHERE_API_KEY = Deno.env.get(\"COHERE_API_KEY\") || \"\";\n const ZAI_API_KEY = Deno.env.get(\"ZAI_API_KEY\") || \"\";\n const DAHL_API_KEY = Deno.env.get(\"DAHL_API_KEY\") || \"\";\n const CLOUDFLARE_ACCOUNT_ID = Deno.env.get(\"CLOUDFLARE_ACCOUNT_ID\") || \"\";\n const CLOUDFLARE_API_TOKEN = Deno.env.get(\"CLOUDFLARE_API_TOKEN\") || \"\";\n const TAVILY_API_KEY = Deno.env.get(\"TAVILY_API_KEY\") || \"\";\n const POLLINATIONS_API_KEY = Deno.env.get(\"POLLINATIONS_API_KEY\") || Deno.env.get(\"POLLINATIONS_KEY\") || \"\";\n const LUMENFALL_API_KEY = Deno.env.get(\"LUMENFALL_API_KEY\") || \"\";\n const HUGGINGFACE_API_KEY = Deno.env.get(\"HUGGINGFACE_API_KEY\") || \"\";\n const APP_SECRET = Deno.env.get(\"APP_SECRET\") || \"\";\n await dbInit();\n\n if (req.method === \"DELETE\" && url.searchParams.get(\"action\") === \"delete\") {\n const ok = await dbDel(url.searchParams.get(\"c\") || \"\");\n return new Response(JSON.stringify({ ok }), { headers: { \"Content-Type\": \"application/json\" } });\n }\n if (req.method === \"POST\" && url.searchParams.get(\"action\") === \"rename\") {\n try {\n const fd = await req.formData();\n const ok = await dbRen(String(fd.get(\"id\") || \"\"), String(fd.get(\"title\") || \"\"));\n return new Response(JSON.stringify({ ok }), { headers: { \"Content-Type\": \"application/json\" } });\n } catch {\n return new Response(JSON.stringify({ ok: false }), { status: 400, headers: { \"Content-Type\": \"application/json\" } });\n }\n }\n if (req.method === \"GET\" && url.searchParams.get(\"action\") === \"export_full\") {\n const cid = url.searchParams.get(\"c\") || \"\";\n if (!cid) return new Response(\"חסר c\", { status: 400 });\n const full = await dbFull(cid);\n const raws = await dbGetRaw(cid);\n let md = \"# שיחה מלאה - v33 LIVE\\n\\n---\\n\\n\";\n const questions = full.filter((m) => m.role === \"user\");\n for (let i = 0; i < questions.length; i++) {\n const q = questions[i];\n md += \"## 👤 שאלה \" + (i + 1) + \"\\n\\n\" + q.content.slice(0, 50000) + \"\\n\\n\";\n const relatedRaw = raws.filter((r) => r.question === q.content);\n for (const r of relatedRaw) md += \"### \" + r.label + \" (\" + r.provider + \")\\n\\n\" + r.answer.slice(0, 50000) + \"\\n\\n\";\n const nextUserIdx = full.findIndex((m, idx) => idx > full.indexOf(q) && m.role === \"user\");\n const finalAns = full.slice(full.indexOf(q) + 1, nextUserIdx === -1 ? undefined : nextUserIdx).find((m) => m.role === \"assistant\");\n if (finalAns) md += \"### 🤖 תשובה סופית\\n\\n\" + finalAns.content.slice(0, 50000) + \"\\n\\n---\\n\\n\";\n }\n return new Response(md, { headers: { \"Content-Type\": \"text/markdown; charset=utf-8\", \"Content-Disposition\": \"attachment; filename=chat-full-\" + Date.now() + \".md\" } });\n }\n if (req.method === \"GET\" && url.searchParams.get(\"action\") === \"export_json\") {\n const cid = url.searchParams.get(\"c\") || \"\";\n if (!cid) return new Response(\"חסר c\", { status: 400 });\n const full = await dbFull(cid);\n const raws = await dbGetRaw(cid);\n return new Response(JSON.stringify({ conversation_id: cid, messages: full, raw_answers: raws, exported_at: Date.now(), version: \"v33\" }, null, 2), { headers: { \"Content-Type\": \"application/json; charset=utf-8\", \"Content-Disposition\": \"attachment; filename=chat-\" + Date.now() + \".json\" } });\n }\n if (req.method === \"GET\" && url.searchParams.get(\"action\") === \"pollinations_models\") {\n const catalog = await fetchPollinationsCatalog();\n return new Response(JSON.stringify(catalog), { headers: { \"Content-Type\": \"application/json\" } });\n }\n\n if (req.method === \"POST\") {\n let q = \"\", cid = \"\", prov = \"\", fileText = \"\";\n let forceSearchParam = \"auto\", forceRecalcParam = false;\n let imgStyle = \"photorealistic\", imgQuality = \"hd\";\n try {\n const fd = await req.formData();\n q = String(fd.get(\"q\") || \"\").trim();\n cid = String(fd.get(\"c\") || \"\");\n prov = String(fd.get(\"p\") || \"\");\n forceSearchParam = String(fd.get(\"forceSearch\") || \"auto\");\n forceRecalcParam = String(fd.get(\"forceRecalc\") || \"\") === \"true\";\n imgStyle = String(fd.get(\"imgStyle\") || \"photorealistic\");\n imgQuality = String(fd.get(\"imgQuality\") || \"hd\");\n for (const [key, val] of fd.entries()) {\n if (key.startsWith(\"file_\") && val instanceof File) {\n try { fileText += \"\\n\\n[קובץ: \" + val.name + \"]\\n\" + (await val.text()).slice(0, MAX_FILE) + \"\\n\"; } catch {}\n }\n }\n } catch {\n return new Response(JSON.stringify({ error: \"בקשה לא תקינה\" }), { status: 400, headers: { \"Content-Type\": \"application/json\" } });\n }\n if (fileText) q = (q + \"\\n\\n--- קבצים ---\\n\" + fileText).slice(0, MAX_Q);\n if (!q) return new Response(JSON.stringify({ error: \"חסרה שאלה\" }), { status: 400, headers: { \"Content-Type\": \"application/json\" } });\n if (q.length > MAX_Q) return new Response(JSON.stringify({ error: \"ארוכה מדי\" }), { status: 400, headers: { \"Content-Type\": \"application/json\" } });\n if (!cid) cid = \"conv_\" + Date.now() + \"_\" + Math.random().toString(36).slice(2, 6);\n if (!checkRateLimit(cid)) return new Response(JSON.stringify({ error: \"יותר מדי בקשות\" }), { status: 429, headers: { \"Content-Type\": \"application/json\" } });\n\n const qNormalized = q.normalize(\"NFC\").trim();\n const trimmedQ = qNormalized.toLowerCase();\n const isImageReq = /^\\/image\\s/i.test(qNormalized) ||\n /^\\/תמונה\\s/i.test(qNormalized) ||\n /^\\/ציור\\s/i.test(qNormalized) ||\n /^(תיצור|צור|תעשה|תייצר|תכין|תאר|תן לי|אני רוצה|אפשר)\\s+(לי\\s+)?(תמונה|ציור|איור)/.test(qNormalized) ||\n /^(צייר לי|צייר|תצייר)\\s/.test(qNormalized) ||\n qNormalized.startsWith(\"תמונה של \") ||\n qNormalized.startsWith(\"🎨\") ||\n /^\\[(photorealistic|anime|3d|fast|draft|hd|ultra)\\]/i.test(qNormalized);\n\n if (isImageReq) {\n let rawPrompt = qNormalized\n .replace(/^\\/image\\s*/i, \"\")\n .replace(/^\\/תמונה\\s*/i, \"\")\n .replace(/^\\/ציור\\s*/i, \"\")\n .replace(/^תמונה\\s+(יפה|יפה מאוד|מגניבה|מקצועית|מרהיבה|מדהימה|חמודה|מושלמת|איכותית)\\s+של\\s+/i, \"\")\n .replace(/^תמונה\\s+(יפה|יפה מאוד|מגניבה|מקצועית|מרהיבה|מדהימה|חמודה|מושלמת|איכותית)\\s+/i, \"\")\n .replace(/^(תיצור|צור|תעשה|תייצר|תכין|תאר|תן לי|אני רוצה|אפשר)\\s+(לי\\s+)?(תמונה|ציור|איור)(\\s+באמצעות\\s+\\S+)?\\s*/i, \"\")\n .replace(/^(צייר לי|צייר|תצייר)\\s+/, \"\")\n .replace(/^תמונה של\\s+/i, \"\")\n .replace(/🎨/g, \"\")\n .trim();\n const allBrackets = [...rawPrompt.matchAll(/\\[([a-z0-9-]+)\\]/gi)].map((m) => m[1].toLowerCase());\n let styleFromQ = imgStyle, qualityFromQ = imgQuality;\n for (const b of allBrackets) {\n if ([\"photorealistic\", \"anime\", \"3d\", \"fast\"].includes(b)) styleFromQ = b;\n if ([\"draft\", \"hd\", \"ultra\"].includes(b)) qualityFromQ = b;\n }\n rawPrompt = rawPrompt.replace(/\\[.*?\\]/g, \"\").trim();\n if (!rawPrompt) rawPrompt = \"beautiful landscape\";\n console.log(\"[IMG] Hebrew prompt:\", rawPrompt);\n const enPrompt = await translateToEnglishSafe(rawPrompt, GEMINI_API_KEY);\n console.log(\"[IMG] English prompt:\", enPrompt);\n const imgRes = await genImage(enPrompt, CLOUDFLARE_ACCOUNT_ID, CLOUDFLARE_API_TOKEN, styleFromQ, qualityFromQ, GEMINI_API_KEY, POLLINATIONS_API_KEY, LUMENFALL_API_KEY, HUGGINGFACE_API_KEY);\n await dbSave(cid, \"user\", q);\n await dbSave(cid, \"assistant\", imgRes.a);\n const t = await generateTitle(q, GEMINI_API_KEY);\n await dbMeta(cid, t);\n return new Response(JSON.stringify({ answer: imgRes.a, raw: [imgRes], memory: 0, search: false, direct: true, totalTokens: 0, conversationId: cid }), { headers: { \"Content-Type\": \"application/json; charset=utf-8\" } });\n }\n\n const qh = qHash(q);\n const skipCache = isNewsQuery(q) || forceSearchParam === \"true\" || forceRecalcParam;\n\n if (!skipCache) {\n const cached = await getCache(qh);\n if (cached) {\n await dbSave(cid, \"user\", q);\n await dbSave(cid, \"assistant\", cached.answer);\n const t = await generateTitle(q, GEMINI_API_KEY);\n await dbMeta(cid, t);\n return new Response(JSON.stringify({ answer: cached.answer, raw: cached.raw, memory: 0, search: false, direct: false, totalTokens: 0, conversationId: cid, cached: true }), { headers: { \"Content-Type\": \"application/json; charset=utf-8\" } });\n }\n }\n\n let enabled = prov ? prov.split(\",\").map((s) => s.trim()).filter((s) => ALL.includes(s)) : ALL.slice();\n const keys: any = {\n gemini: GEMINI_API_KEY, groq: GROQ_API_KEY, groq2: GROQ_API_KEY_2, deepseek: DEEPSEEK_API_KEY,\n cerebras: CEREBRAS_API_KEY, nvidia: NVIDIA_API_KEY, mistral: MISTRAL_API_KEY,\n openrouter: OPENROUTER_API_KEY, cohere: COHERE_API_KEY, zai: ZAI_API_KEY, dahl: DAHL_API_KEY,\n llm7: \"unused\",\n zen: \"public\",\n poll: \"no-key\",\n keyless: \"not-needed\",\n };\n enabled = enabled.filter((p) => keys[p] || (p === \"groq\" && (keys.groq || keys.groq2)));\n if (enabled.length === 0) return new Response(JSON.stringify({ error: \"אין מנועים\" }), { status: 400, headers: { \"Content-Type\": \"application/json\" } });\n\n const stream = new ReadableStream({\n async start(controller) {\n const enc = new TextEncoder();\n const send = (obj: any) => {\n try { controller.enqueue(enc.encode(JSON.stringify(obj) + \"\\n\")); } catch {}\n };\n\n try {\n send({\n type: \"start\",\n total: enabled.length,\n providers: enabled.map((pk) => ({ pk, label: PL[pk] })),\n });\n\n const hist = await dbHist(cid);\n let shouldSearch = false;\n if (TAVILY_API_KEY && !isMetaQuestion(q)) {\n if (forceSearchParam === \"true\") shouldSearch = true;\n else if (forceSearchParam === \"false\") shouldSearch = isNewsQuery(q);\n else shouldSearch = isNewsQuery(q) || q.length > 10;\n }\n\n if (shouldSearch) {\n send({ type: \"status\", message: \"🔍 מחפש באינטרנט...\" });\n }\n\n const searchCtx = shouldSearch ? await wt(webSearch(q, TAVILY_API_KEY), 7000, \"\") : \"\";\n const fullPrompt = ctx(hist, q) + (searchCtx ? \"\\n\\nWeb search results:\\n\" + searchCtx : \"\");\n\n const round1 = await Promise.all(enabled.map(async (pk) => {\n const r = await askProvider(fullPrompt, keys, pk);\n send({\n type: \"done\",\n pk,\n label: r.p,\n ok: !!(r.a && !r.e),\n ms: r.ms || 0,\n e: r.e || \"\",\n preview: r.a ? r.a.slice(0, 100) : \"\",\n });\n return r;\n }));\n\n const valid = round1.filter((a) => a.a && a.a.length > 20 && !a.a.includes(\"המערכת פותחה על ידי מוטי כץ\"));\n const fb = valid.length > 0 ? valid : round1.filter((a) => a.a && a.a.length > 0);\n\n let finalAnswer = \"\";\n\n if (fb.length === 0) {\n send({ type: \"status\", message: \"⚠ כל המנועים נכשלו, מנסה שופט בלבד...\" });\n if (searchCtx) {\n const jpF = buildJP(q, [], hist, searchCtx);\n const jrF = await runJudge(jpF, CLOUDFLARE_ACCOUNT_ID, CLOUDFLARE_API_TOKEN, GEMINI_API_KEY);\n const parsed = parseJP(jrF || \"\");\n if (validateAnswer(parsed.answer)) {\n finalAnswer = parsed.answer;\n } else {\n finalAnswer = \"מצטער, לא הצלחתי להפיק תשובה מהמנועים. נסה שוב בעוד רגע.\";\n }\n } else {\n finalAnswer = \"כל המנועים נכשלו כרגע. נסה שוב בעוד רגע.\";\n }\n } else if (fb.length === 1) {\n send({ type: \"status\", message: \"🎯 מנוע אחד הצליח, משתמש בתשובתו...\" });\n finalAnswer = fb[0].a;\n } else {\n send({ type: \"judging\", count: fb.length, total: round1.length });\n const jp = buildJP(q, round1, hist, searchCtx);\n const jr = await runJudge(jp, CLOUDFLARE_ACCOUNT_ID, CLOUDFLARE_API_TOKEN, GEMINI_API_KEY);\n const parsed = parseJP(jr || \"\");\n if (validateAnswer(parsed.answer)) {\n finalAnswer = parsed.answer;\n } else {\n finalAnswer = fb[0].a;\n }\n }\n\n await dbSave(cid, \"user\", q);\n await dbSaveRaw(cid, q, round1.filter((r) => r.a));\n await dbSave(cid, \"assistant\", finalAnswer);\n const title = await generateTitle(q, GEMINI_API_KEY);\n await dbMeta(cid, title);\n if (!isNewsQuery(q)) await setCache(qh, q, finalAnswer, round1);\n const totalTokens = round1.reduce((s, a) => s + (a.ti || 0) + (a.to || 0), 0);\n\n send({\n type: \"final\",\n answer: finalAnswer,\n raw: round1,\n memory: hist.length,\n search: !!searchCtx,\n searchContent: searchCtx,\n direct: fb.length === 1,\n totalTokens,\n conversationId: cid,\n cached: false,\n });\n } catch (e) {\n send({ type: \"final\", answer: \"⚠ שגיאה: \" + String(e).slice(0, 200), raw: [], memory: 0, search: false, direct: false, totalTokens: 0, conversationId: cid, cached: false });\n } finally {\n try { controller.close(); } catch {}\n }\n },\n });\n\n return new Response(stream, {\n headers: {\n \"Content-Type\": \"application/x-ndjson; charset=utf-8\",\n \"Cache-Control\": \"no-store\",\n \"X-Accel-Buffering\": \"no\",\n },\n });\n }\n\n const cid = url.searchParams.get(\"c\") || \"\";\n const convs = await dbList();\n const msgs = cid ? await dbFull(cid) : [];\n const health = await dbGetHealth();\n const avail: string[] = [];\n if (GEMINI_API_KEY) avail.push(\"gemini\");\n if (GROQ_API_KEY || GROQ_API_KEY_2) avail.push(\"groq\");\n if (DEEPSEEK_API_KEY) avail.push(\"deepseek\");\n if (CEREBRAS_API_KEY) avail.push(\"cerebras\");\n if (NVIDIA_API_KEY) avail.push(\"nvidia\");\n if (MISTRAL_API_KEY) avail.push(\"mistral\");\n if (OPENROUTER_API_KEY) avail.push(\"openrouter\");\n if (COHERE_API_KEY) avail.push(\"cohere\");\n if (ZAI_API_KEY) avail.push(\"zai\");\n if (DAHL_API_KEY) avail.push(\"dahl\");\n avail.push(\"llm7\");\n avail.push(\"zen\");\n avail.push(\"poll\");\n avail.push(\"keyless\");\n return renderPage(cid, convs, msgs, avail, health);\n}\n\n// === END OF FILE ===",[168],"pulls","branches",1,"forks","envVars",{"_182":184,"_185":26},"connections",{"_182":26,"_183":26},"count","unfilledCount",16,"emptyCount",{"_33":167,"_13":168,"_62":91,"_43":-5,"_35":36,"_158":118,"_160":78,"_37":192,"_39":170,"_64":171,"_163":172,"_66":24,"_165":-5,"_189":193,"_17":194},{"_33":155,"_13":156,"_62":157,"_43":-5,"_35":36,"_158":159,"_160":78,"_37":188,"_39":162,"_64":65,"_163":164,"_66":24,"_165":-5,"_189":190,"_17":191},["D",1789575102592],"children",[],[156],["D",1789579732258],[],[168],["D",1789494982975],["D",1789522252250],{"_11":12,"_132":69},{"_11":12,"_127":128,"_68":69,"_30":67,"_129":130,"_131":24},{"_39":118,"_13":16,"_33":36,"_119":80,"_120":-5,"_121":-5,"_122":-5,"_123":-5,"_124":-5,"_125":14,"_126":12},"ctx",{"_52":-5,"_150":233,"_234":235,"_236":237,"_238":24,"_239":24,"_240":24,"_241":78,"_242":-5,"_243":-7,"_244":-7,"_245":26,"_246":24,"_247":24,"_248":-5,"_249":24,"_250":24,"_251":24,"_252":-5},"ENV",{"_204":205,"_206":207,"_208":16,"_209":210,"_211":212,"_213":214,"_215":216,"_217":218,"_219":220,"_221":222,"_223":224,"_225":226,"_227":228,"_229":230,"_231":232},"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"]

🍌 ננו-בננה ULTRA

History · new-file-6277.ts