Semantic Search

\n \n

Uses OpenRouter embeddings + SQLite vector search (assumed).

\n \n `,\n{ headers: { \"Content-Type\": \"text/html; charset=utf-8\" } } ); }\n\n// API endpoint: /search?q=... if (req.method === \"GET\" && url.pathname ===\n\"/search\") { const q = url.searchParams.get(\"q\") ?? \"\"; const k =\nNumber(url.searchParams.get(\"k\") ?? \"10\");\n\n if (!q.trim()) {\n return new Response(JSON.stringify({ error: \"Missing q\" }), {\n status: 400,\n headers: { \"Content-Type\": \"application/json\" },\n });\n }\n\n const results = await semanticSearch(q, Number.isFinite(k) ? k : 10);\n\n // If you want HTML results when opened in browser:\n const accept = req.headers.get(\"accept\") ?? \"\";\n if (accept.includes(\"text/html\")) {\n const items = results.map((r: any) => `\n

${escapeHtml(String(r.doc_title))}

doc: ${escapeHtml(String(r.doc_id))} · chunk ${r.chunk_index}

${escapeHtml(String(r.chunk_text))}

\n `).join(\"\");\n\n return new Response(\n `\n \n Results\n \n ← back\n

Results for: ${escapeHtml(q)}

\n ${items || \"

No results

\"}\n \n \n `,\n { headers: { \"Content-Type\": \"text/html; charset=utf-8\" } }\n );\n }\n\n // JSON by default\n return new Response(JSON.stringify({ q, results }), {\n headers: { \"Content-Type\": \"application/json\" },\n });\n\n}\n\nreturn new Response(\"Not found\", { status: 404 }); }\n\nfunction escapeHtml(s: string) { return s .replaceAll(\"&\", \"&\")\n.replaceAll(\"<\", \"<\") .replaceAll(\">\", \">\") .replaceAll('\"', \""\")\n.replaceAll(\"'\", \"'\"); }\n\n9. Add documents (two easy ways) Option A: Hardcode a “seed docs” function\n (quickest for noobs)\n\nCreate a val seedDocs:\n\nimport { setupDb } from \"./dbSetup.ts\"; import { upsertDoc } from \"./docs.ts\";\nimport { reindexDoc } from \"./indexer.ts\";\n\nexport async function seedDocs() { await setupDb();\n\nconst docs = [ { id: \"doc-hello\", title: \"Hello Doc\", content: \"Hello world.\nThis is a test document about dogs, cats, and APIs.\", }, { id: \"doc-webhooks\",\ntitle: \"Webhook Notes\", content: \"To verify a webhook signature, compute HMAC of\nthe payload and compare.\", }, ];\n\nfor (const d of docs) { await upsertDoc(d.id, d.title, d.content); await\nreindexDoc(d.id); }\n\nreturn { ok: true, count: docs.length }; }\n\nRun seedDocs(). Then open your HTTP val and search.\n\nOption B: Create an HTTP “ingest” endpoint (more practical)\n\nAdd to your HTTP handler:\n\nPOST /ingest with {id,title,content}\n\nupsert + reindex\n\n(Ask and I’ll paste the exact code, but the above already gives you the\nprimitives.)\n\n10. Add a Cron val (optional, recommended)\n\nIf your documents change elsewhere, you can reindex on a schedule.\n\nCron val example:\n\nEvery hour:\n\nselect docs updated recently\n\nreindex them\n\nimport sqlite from \"https://esm.town/v/std/sqlite/main.ts\"; import { setupDb }\nfrom \"./dbSetup.ts\"; import { reindexDoc } from \"./indexer.ts\";\n\nexport default async function cron() { await setupDb();\n\n// Reindex docs updated in the last 2 hours (example) const since = Date.now() -\n2 * 60 * 60 * 1000;\n\nconst res = await sqlite.execute( `SELECT id FROM docs WHERE updated_at >= ?`,\n[since] );\n\nfor (const row of res.rows ?? []) { await reindexDoc(String(row.id)); }\n\nreturn { ok: true, reindexed: res.rows?.length ?? 0 }; }\n\n11. “Hybrid search” upgrade (semantic + exact keywords)\n\nOnce you have semantic working, add lexical search so exact identifiers/words\nwin when appropriate.\n\n11.1 Create an FTS table await\nsqlite.execute(`CREATE VIRTUAL TABLE IF NOT EXISTS chunk_fts\n USING fts5(chunk_id, chunk_text);`);\n\n11.2 Keep it in sync when indexing\n\nAfter inserting a chunk, also upsert into FTS:\n\nawait sqlite.execute(\n`INSERT INTO chunk_fts (chunk_id, chunk_text) VALUES (?, ?)`, [chunkId,\nchunkText] );\n\nIf chunk ids can be reinserted, you may prefer:\n\ndelete old FTS rows for doc first, then insert fresh\n\n11.3 Hybrid query\n\nGet top 50 semantic chunks\n\nGet top 50 lexical chunks\n\nMerge + dedupe in TypeScript\n\nSort with a simple scoring heuristic (semantic rank + lexical rank)\n\nThis is extremely effective and beginner-friendly.\n\n12. Common pitfalls + fixes Pitfall A: “Vector column dimension mismatch”\n\nSymptoms:\n\ninsert fails\n\nor vector_top_k fails\n\nFix:\n\nlog embedding length using testEmbeddingDim()\n\nchange F32_BLOB(1024) to the real number\n\nrebuild DB table (or create a new table)\n\nPitfall B: “SQL functions not found” (vector_top_k, libsql_vector_idx)\n\nThis means the assumed native vector layer isn’t enabled.\n\nFor this guide, we assume it is. If it isn’t, the fallback paths are:\n\nstore embeddings as JSON blobs, do brute-force cosine in JS (slow but works)\n\nuse Postgres + pgvector externally (like the 2024 blog post)\n\nuse a hosted vector DB\n\nBut don’t mix those into this manual yet—you asked to assume it works.\n\nPitfall C: Timeouts / rate limits from OpenRouter\n\nFixes:\n\nreduce batch size (e.g., 16)\n\nadd a small delay between batches\n\ncache embeddings using content_hash so you don’t recompute unchanged chunks\n\nPitfall D: High cost\n\nEven if embeddings are cheap, big docs + too many chunks adds up.\n\nFixes:\n\nchunk bigger (1200–1500 chars)\n\nonly embed changed docs\n\nadd a “max docs indexed” limit in early prototypes\n\n13. Minimal checklist (copy/paste workflow)\n\nAdd env var: OPENROUTER_API_KEY\n\nRun testEmbeddingDim() → note length D\n\nSet schema F32_BLOB(D)\n\nRun setupDb()\n\nAdd docs (seed or ingest)\n\nRun reindexDoc(docId) for each doc\n\nHit /search?q=...",[],[183],{"_105":193,"_97":194,"_163":195,"_165":-5,"_166":148,"_167":196,"_169":22,"_107":197,"_101":187,"_172":188,"_174":198,"_176":22,"_177":-5,"_178":199,"_100":200},"96bb406c-ffdc-11f0-acb7-42dde27851f2","review.md","8d17d5d4-ffdc-11f0-a811-42dde27851f2",10,["D",1769998333512],"No-holds-barred take (based on the portion you pasted — it cuts off\nmid-function, so I’m reviewing what’s visible + the implied next steps).\n\n---\n\n## 1) The good parts (you’re on the right track)\n\n### ✅ You’re testing the _right_ primitives\n\nYou’re explicitly validating the exact libSQL/Turso “native vectors” surface\narea Val Town would expose:\n\n- `F32_BLOB(D)` vector column types\n- `vector(...)` constructor\n- `libsql_vector_idx(...)` index marker\n- `vector_top_k(...)` query function\n\nThat’s exactly the feature set Turso documents for vector search.\n([docs.turso.tech][1])\n\n### ✅ You’re checking embedding dimensionality (at least printing it)\n\nPrinting `embedding.length` is the first sanity check that prevents 90% of “why\nwon’t inserts / index queries work?” issues.\n\n### ✅ Your SQL literal conversion is conceptually correct\n\nIn libSQL, `vector` is an alias conversion function (effectively `vector32`) and\nit accepts a JSON array string. ([docs.turso.tech][1])\n\n---\n\n## 2) The big problems / likely breakpoints\n\n### 🚨 Problem A: You hardcode `EMBEDDING_DIM = 4096` but you never _enforce_ it\n\nRight now you _print_ length, but you don’t fail fast if the returned length\ndiffers.\n\nWhy this matters:\n\n- libSQL requires the query vector to have the **same type and dimensionality**\n as the column/index. If your table is `F32_BLOB(4096)` and the API returns\n 3072 or 1024, you’re dead. ([docs.turso.tech][1])\n\n**Fix:** assert the dimension _before_ any DB work.\n\n```ts\nfunction assertDim(vec: number[], expected: number) {\n if (vec.length !== expected) {\n throw new Error(\n `Embedding dim mismatch: got ${vec.length}, expected ${expected}`,\n );\n }\n}\n```\n\nThen in `testEmbeddingAPI()`:\n\n```ts\nassertDim(embedding, EMBEDDING_DIM);\n```\n\nAlso: Qwen3-Embedding supports up to 4096 dims (and sometimes providers always\nreturn 4096). ([Hugging Face][2])\n\n---\n\n### 🚨 Problem B: `toFixed(8)` is not helping you, and it can hurt you\n\nYou’re rounding every component to 8 decimals. That is:\n\n- wasted CPU\n- makes the SQL string bigger than it needs to be (because you force a fixed\n representation)\n- introduces quantization noise you didn’t ask for\n\nlibSQL will store as float32 anyway; you don’t need to pre-round.\n\n**Fix:** remove rounding completely (or keep minimal rounding only for debugging\nprints).\n\n---\n\n### 🚨 Problem C: Building a 4096-float SQL literal is a “string bazooka”\n\nA 4096-dim embedding as JSON text is **huge**. You’re generating a SQL statement\nthat contains a massive `vector('[...]')` blob.\n\nThis can bite you in multiple ways:\n\n- request payload size to the DB endpoint\n- SQL parsing overhead\n- log output bloat if anything errors and prints the query\n- harder to parameterize safely\n\nTurso’s docs explicitly support converting a **text JSON array** to the target\nvector type via conversion functions like `vector32(...)`.\n([docs.turso.tech][1])\n\n**Much better fix:** parameterize the JSON and wrap it in `vector32(?)`:\n\n```ts\nconst vecJson = JSON.stringify(embedding);\n\nawait sqlite.execute({\n sql: `INSERT INTO t(embedding) VALUES (vector32(?))`,\n args: [vecJson],\n});\n```\n\nThis keeps SQL smaller, avoids injection-y patterns, and is closer to the\ndocumented usage model. ([docs.turso.tech][1])\n\n---\n\n### 🚨 Problem D: You’re importing std/sqlite in a nonstandard way\n\nYou used:\n\n```ts\nimport sqlite from \"https://esm.town/v/std/sqlite@14-main/main.ts\";\n```\n\nVal Town’s own std/sqlite README shows:\n\n```ts\nimport sqlite from \"https://esm.town/v/std/sqlite/main.ts\";\n```\n\nSo unless you _need_ that exact tag/branch, you’re taking on avoidable breakage\nrisk. ([Val Town][3])\n\n**Fix (simple):**\n\n```ts\nimport sqlite from \"https://esm.town/v/std/sqlite/main.ts\";\n```\n\nIf you want pinning, pin a commit/version you control—but don’t invent tag\nshapes unless you know they exist.\n\n---\n\n### 🚨 Problem E: You’re not testing the _actual_ vector mechanics libSQL cares about\n\nThe real “gotchas” are:\n\n1. **Conversion function/type match** Your table will likely be `F32_BLOB(D)`.\n Your inserted values must be float32 vectors. `vector(...)` is alias for\n float32 conversion, which is good. ([docs.turso.tech][1])\n\n2. **Index query requires same type + same dimensionality** `vector_top_k` will\n fail if the query vector differs in either. ([docs.turso.tech][1])\n\nYour script _should_ include a test that:\n\n- creates the table with `F32_BLOB(4096)`\n- creates the index with `libsql_vector_idx(embedding, 'metric=cosine')`\n- inserts at least 2 rows\n- runs `vector_top_k(...)` and joins back to the table\n\nTurso’s docs + blog show this exact flow. ([turso.tech][4])\n\n---\n\n## 3) Performance + cost reality check (the stuff that will sting later)\n\n### 4096 dims is allowed, but it’s chunky\n\nTurso docs: libSQL can operate up to **65536 dimensions**. So 4096 is fine.\n([docs.turso.tech][1])\n\nBut storage math: float32 = 4 bytes.\n\n- 4096 × 4 = **16384 bytes (~16 KB) per row**, _just for the embedding_, before\n index overhead.\n\nIf you chunk docs and store many rows, that balloons fast.\n\n**Hard truth:** Unless you truly need 4096, you’ll often be better with a\nsmaller vector (common sweet spots: 768–1536) because your recall gain from\nbigger dims is frequently marginal vs your cost/index size. (Not always, but\noften.)\n\n---\n\n## 4) Security / reliability issues\n\n### API call reliability\n\nYour embedding call has:\n\n- no timeout\n- no retry/backoff\n- no provider error normalization\n- no batching guardrails\n\nNebius will occasionally spike latency or return transient errors—this will make\nindexing brittle.\n\n**Fix:** add an `AbortController` timeout + small retry.\n\n### Leak risk\n\nDon’t log the entire embedding ever (you currently only print first 5 values —\ngood). Keep it that way.\n\n---\n\n## 5) Concrete code upgrades (drop-in improvements)\n\n### 5.1 Replace `toVectorSqlLiteral` with parameterized `vector32(?)`\n\n```ts\nfunction vectorArgs(embedding: number[]) {\n // no rounding; keep raw floats\n return JSON.stringify(embedding);\n}\n\n// usage:\nawait sqlite.execute({\n sql: `INSERT INTO test_vectors(id, embedding)\n VALUES (?, vector32(?))`,\n args: [\"row1\", vectorArgs(embedding)],\n});\n```\n\nThis matches Turso’s documented “text JSON array → vector32 conversion” path.\n([docs.turso.tech][1])\n\n### 5.2 Enforce dimension immediately\n\n```ts\nfunction assertDim(vec: number[], expected: number) {\n if (vec.length !== expected) {\n throw new Error(\n `Embedding dim mismatch: got ${vec.length}, expected ${expected}`,\n );\n }\n}\n```\n\n### 5.3 Use the canonical std/sqlite import\n\n```ts\nimport sqlite from \"https://esm.town/v/std/sqlite/main.ts\";\n```\n\nAs documented by Val Town’s std/sqlite README + docs. ([Val Town][3])\n\n### 5.4 Your DB test should explicitly validate `vector_top_k` semantics\n\nTurso docs explain: `vector_top_k` returns `ROWID` (or `PRIMARY KEY` for WITHOUT\nROWID tables), and you usually join back to the base table.\n([docs.turso.tech][1])\n\nSo your test should look like:\n\n```ts\nawait sqlite.execute(`DROP TABLE IF EXISTS test_vectors`);\nawait sqlite.execute(`\n CREATE TABLE test_vectors (\n id TEXT PRIMARY KEY,\n embedding F32_BLOB(4096)\n )\n`);\nawait sqlite.execute(`\n CREATE INDEX test_vectors_idx\n ON test_vectors (libsql_vector_idx(embedding, 'metric=cosine'))\n`);\n\nconst [e1] = await embedText(\"dogs bark\");\nconst [e2] = await embedText(\"github api client\");\nassertDim(e1, 4096);\nassertDim(e2, 4096);\n\nawait sqlite.execute({\n sql: `INSERT INTO test_vectors(id, embedding) VALUES (?, vector32(?))`,\n args: [\"a\", JSON.stringify(e1)],\n});\nawait sqlite.execute({\n sql: `INSERT INTO test_vectors(id, embedding) VALUES (?, vector32(?))`,\n args: [\"b\", JSON.stringify(e2)],\n});\n\nconst [q] = await embedText(\"discord bot integration\");\nassertDim(q, 4096);\n\nconst res = await sqlite.execute({\n sql: `\n SELECT tv.id\n FROM vector_top_k('test_vectors_idx', vector32(?), 2) v\n JOIN test_vectors tv ON tv.rowid = v.id\n `,\n args: [JSON.stringify(q)],\n});\n\nconsole.log(res.rows);\n```\n\nThat test is aligned with how Turso describes indexing/querying.\n([docs.turso.tech][1])\n\n---\n\n## 6) Verdict (no sugar-coating)\n\n**Your direction is correct, but your current implementation style (giant SQL\nstring literals + unused dimension constant + nonstandard sqlite import) is\nexactly the kind of thing that “works once” and then becomes flaky and painful\nto productionize.**\n\nIf you implement just these 3 changes:\n\n1. **parameterize vectors with `vector32(?)`**\n2. **assert dimensions**\n3. **use canonical `std/sqlite/main.ts` import**\n\n…your script goes from “prototype that might lie to you” to “reliable foundation\ntest”.\n\n---\n\nIf you paste the rest (especially the SQLite tests you wrote after the cutoff),\nI’ll do an equally ruthless line-by-line review of the actual table/index/query\nparts—because that’s where most people accidentally test the wrong thing and\nthink vectors “don’t work” when it’s really a join/type mismatch.\n\n[1]: https://docs.turso.tech/features/ai-and-embeddings \"AI & Embeddings - Turso\"\n[2]: https://huggingface.co/Qwen/Qwen3-Embedding-8B?utm_source=chatgpt.com \"Qwen/Qwen3-Embedding-8B\"\n[3]: https://www.val.town/v/std/sqlite \"std/sqlite | Val Town\"\n[4]: https://turso.tech/blog/approximate-nearest-neighbor-search-with-diskann-in-libsql \"Approximate nearest neighbor search with DiskANN in libSQL\"",[],[194],{"_105":202,"_97":203,"_163":204,"_165":-5,"_166":148,"_167":102,"_169":22,"_107":205,"_101":33,"_172":206,"_174":207,"_176":22,"_177":-5,"_178":208,"_100":209},"b5aec216-ffdf-11f0-9ad7-42dde27851f2","test_nebius_full.ts","b5aec306-ffdf-11f0-9ad7-42dde27851f2",["D",1769999673928],"http","/**\n * Full Integration Test: Nebius API + SQLite 4096-dim Vectors\n * \n * Tests:\n * 1. Nebius API embedding generation\n * 2. Dimension verification (should be 4096)\n * 3. Insert 4096-dim vectors into SQLite\n * 4. Vector search with real embeddings\n * \n * Returns JSON - takes ~5-10 seconds\n */\n\nimport sqlite from \"https://esm.town/v/std/sqlite/main.ts\";\n\nconst NEBIUS_URL = \"https://api.tokenfactory.nebius.com/v1/embeddings\";\nconst EMBEDDING_MODEL = \"Qwen/Qwen3-Embedding-8B\";\nconst EMBEDDING_DIM = 4096;\n\ninterface NebiusEmbeddingResponse {\n object: string;\n data: Array<{\n object: string;\n index: number;\n embedding: number[];\n }>;\n model: string;\n usage: {\n prompt_tokens: number;\n total_tokens: number;\n };\n}\n\nasync function embedText(input: string | string[]): Promise {\n const apiKey = Deno.env.get(\"NEBIUS_API_KEY\");\n if (!apiKey) throw new Error(\"Missing NEBIUS_API_KEY env var\");\n\n const controller = new AbortController();\n const timeoutId = setTimeout(() => controller.abort(), 30000);\n\n try {\n const res = await fetch(NEBIUS_URL, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${apiKey}`,\n \"Content-Type\": \"application/json\",\n \"Accept\": \"*/*\",\n },\n body: JSON.stringify({\n model: EMBEDDING_MODEL,\n input: input,\n }),\n signal: controller.signal,\n });\n\n if (!res.ok) {\n const errText = await res.text();\n throw new Error(`Nebius error ${res.status}: ${errText}`);\n }\n\n const json = (await res.json()) as NebiusEmbeddingResponse;\n return json.data.map((d) => d.embedding);\n } finally {\n clearTimeout(timeoutId);\n }\n}\n\nexport default async function handler(req: Request): Promise {\n const results: Record = {};\n const startTime = Date.now();\n\n // Test 1: Get embedding from Nebius\n let testEmbeddings: number[][] = [];\n try {\n testEmbeddings = await embedText([\"dogs bark\", \"cats meow\", \"github api\"]);\n results[\"1_nebius_api\"] = { \n success: true, \n data: { \n count: testEmbeddings.length,\n dimensions: testEmbeddings.map(e => e.length),\n }\n };\n } catch (e) {\n results[\"1_nebius_api\"] = { success: false, error: String(e) };\n return Response.json({\n summary: \"Failed at Nebius API call\",\n vector_support: false,\n results,\n elapsed_ms: Date.now() - startTime,\n });\n }\n\n // Test 2: Verify dimensions\n try {\n for (const emb of testEmbeddings) {\n if (emb.length !== EMBEDDING_DIM) {\n throw new Error(`Expected ${EMBEDDING_DIM} dims, got ${emb.length}`);\n }\n }\n results[\"2_dimension_check\"] = { success: true, data: `All ${EMBEDDING_DIM} dims` };\n } catch (e) {\n results[\"2_dimension_check\"] = { success: false, error: String(e) };\n return Response.json({\n summary: \"Failed at dimension check\",\n vector_support: false,\n results,\n elapsed_ms: Date.now() - startTime,\n });\n }\n\n // Test 3: Create table with F32_BLOB(4096)\n try {\n await sqlite.execute(`DROP TABLE IF EXISTS _test_nebius_full`);\n await sqlite.execute(`\n CREATE TABLE _test_nebius_full (\n id TEXT PRIMARY KEY,\n content TEXT,\n emb F32_BLOB(${EMBEDDING_DIM})\n )\n `);\n await sqlite.execute(`\n CREATE INDEX _test_nebius_idx \n ON _test_nebius_full (libsql_vector_idx(emb, 'metric=cosine'))\n `);\n results[\"3_create_table\"] = { success: true };\n } catch (e) {\n results[\"3_create_table\"] = { success: false, error: String(e) };\n return Response.json({\n summary: \"Failed to create table\",\n vector_support: false,\n results,\n elapsed_ms: Date.now() - startTime,\n });\n }\n\n // Test 4: Insert real 4096-dim embeddings\n try {\n const docs = [\n { id: \"doc1\", content: \"dogs bark\", emb: testEmbeddings[0] },\n { id: \"doc2\", content: \"cats meow\", emb: testEmbeddings[1] },\n { id: \"doc3\", content: \"github api\", emb: testEmbeddings[2] },\n ];\n\n for (const doc of docs) {\n await sqlite.execute({\n sql: `INSERT INTO _test_nebius_full (id, content, emb) VALUES (?, ?, vector32(?))`,\n args: [doc.id, doc.content, JSON.stringify(doc.emb)],\n });\n }\n \n results[\"4_insert_4096\"] = { success: true, data: `Inserted ${docs.length} docs` };\n } catch (e) {\n results[\"4_insert_4096\"] = { success: false, error: String(e) };\n return Response.json({\n summary: \"Failed to insert 4096-dim vectors\",\n vector_support: false,\n results,\n elapsed_ms: Date.now() - startTime,\n });\n }\n\n // Test 5: Search with real embedding\n try {\n const [queryEmb] = await embedText(\"furry pets\");\n \n if (queryEmb.length !== EMBEDDING_DIM) {\n throw new Error(`Query embedding wrong size: ${queryEmb.length}`);\n }\n\n const searchResult = await sqlite.execute({\n sql: `\n SELECT t.id, t.content\n FROM vector_top_k('_test_nebius_idx', vector32(?), 2) v\n JOIN _test_nebius_full t ON t.rowid = v.id\n `,\n args: [JSON.stringify(queryEmb)],\n });\n\n results[\"5_vector_search\"] = { \n success: true, \n data: {\n query: \"furry pets\",\n results: searchResult.rows,\n }\n };\n } catch (e) {\n results[\"5_vector_search\"] = { success: false, error: String(e) };\n }\n\n // Cleanup\n try {\n await sqlite.execute(`DROP TABLE IF EXISTS _test_nebius_full`);\n } catch (_) {\n // ignore\n }\n\n const passed = Object.values(results).filter(r => r.success).length;\n const total = Object.keys(results).length;\n const elapsed_ms = Date.now() - startTime;\n\n return Response.json({\n summary: `${passed}/${total} tests passed in ${elapsed_ms}ms`,\n vector_support: passed === total,\n results,\n elapsed_ms,\n }, {\n headers: { \"Content-Type\": \"application/json\" },\n });\n}\n",[],[203],{"_105":211,"_97":212,"_163":213,"_165":-5,"_166":148,"_167":214,"_169":22,"_107":215,"_101":33,"_172":173,"_174":216,"_176":22,"_177":-5,"_178":217,"_100":218},"e5970bca-ffdd-11f0-b73e-42dde27851f2","test_simple.ts","e5970cba-ffdd-11f0-b73e-42dde27851f2",12,["D",1769998895312],"/**\n * Simple diagnostic test - logs immediately to see if anything works\n */\n\nconsole.log(\"=== SIMPLE TEST START ===\");\nconsole.log(\"Timestamp:\", new Date().toISOString());\n\n// Test 1: Basic logging works\nconsole.log(\"✅ Step 1: console.log works\");\n\n// Test 2: Check env var exists (don't log the value!)\nconst hasNebiusKey = !!Deno.env.get(\"NEBIUS_API_KEY\");\nconsole.log(\"✅ Step 2: NEBIUS_API_KEY exists:\", hasNebiusKey);\n\n// Test 3: Try importing sqlite\nconsole.log(\"⏳ Step 3: Importing sqlite...\");\ntry {\n const sqlite = await import(\"https://esm.town/v/std/sqlite/main.ts\");\n console.log(\"✅ Step 3: sqlite imported successfully\");\n console.log(\" sqlite.execute exists:\", typeof sqlite.default?.execute);\n} catch (e) {\n console.log(\"❌ Step 3: sqlite import FAILED:\", e instanceof Error ? e.message : String(e));\n}\n\n// Test 4: Try a simple sqlite query\nconsole.log(\"⏳ Step 4: Testing sqlite.execute...\");\ntry {\n const sqlite = (await import(\"https://esm.town/v/std/sqlite/main.ts\")).default;\n const result = await sqlite.execute(\"SELECT 1 as test\");\n console.log(\"✅ Step 4: sqlite.execute works! Result:\", JSON.stringify(result.rows));\n} catch (e) {\n console.log(\"❌ Step 4: sqlite.execute FAILED:\", e instanceof Error ? e.message : String(e));\n}\n\n// Test 5: Try creating a table with F32_BLOB\nconsole.log(\"⏳ Step 5: Testing F32_BLOB...\");\ntry {\n const sqlite = (await import(\"https://esm.town/v/std/sqlite/main.ts\")).default;\n await sqlite.execute(\"DROP TABLE IF EXISTS _diag_test\");\n await sqlite.execute(\"CREATE TABLE _diag_test (id INTEGER, emb F32_BLOB(3))\");\n console.log(\"✅ Step 5: F32_BLOB column created!\");\n await sqlite.execute(\"DROP TABLE IF EXISTS _diag_test\");\n} catch (e) {\n console.log(\"❌ Step 5: F32_BLOB FAILED:\", e instanceof Error ? e.message : String(e));\n}\n\n// Test 6: Try Nebius API\nconsole.log(\"⏳ Step 6: Testing Nebius API...\");\ntry {\n const apiKey = Deno.env.get(\"NEBIUS_API_KEY\");\n if (!apiKey) throw new Error(\"No API key\");\n \n const res = await fetch(\"https://api.tokenfactory.nebius.com/v1/embeddings\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${apiKey}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n model: \"Qwen/Qwen3-Embedding-8B\",\n input: \"test\",\n }),\n });\n \n if (!res.ok) {\n const text = await res.text();\n throw new Error(`HTTP ${res.status}: ${text.slice(0, 200)}`);\n }\n \n const json = await res.json();\n const dim = json.data?.[0]?.embedding?.length;\n console.log(\"✅ Step 6: Nebius API works! Embedding dimension:\", dim);\n} catch (e) {\n console.log(\"❌ Step 6: Nebius API FAILED:\", e instanceof Error ? e.message : String(e));\n}\n\nconsole.log(\"=== SIMPLE TEST COMPLETE ===\");\n",[],[212],{"_105":220,"_97":221,"_163":222,"_165":-5,"_166":148,"_167":223,"_169":22,"_107":224,"_101":33,"_172":206,"_174":225,"_176":22,"_177":-5,"_178":226,"_100":227},"316a9d68-ffdf-11f0-bfce-42dde27851f2","test_sqlite_vector.ts","316a9e08-ffdf-11f0-bfce-42dde27851f2",13,["D",1769999452023],"/**\n * Minimal SQLite Vector Support Test\n * \n * Tests ONLY SQLite features - NO external API calls\n * Returns JSON so you can see results in browser\n * \n * Should complete in < 1 second\n */\n\nimport sqlite from \"https://esm.town/v/std/sqlite/main.ts\";\n\nexport default async function handler(req: Request): Promise {\n const results: Record = {};\n \n // Test 1: Can we create F32_BLOB column?\n try {\n await sqlite.execute(`DROP TABLE IF EXISTS _test_f32blob`);\n await sqlite.execute(`CREATE TABLE _test_f32blob (id TEXT PRIMARY KEY, emb F32_BLOB(3))`);\n results[\"1_f32_blob_column\"] = { success: true };\n } catch (e) {\n results[\"1_f32_blob_column\"] = { success: false, error: String(e) };\n }\n\n // Test 2: Can we insert with vector32()?\n try {\n await sqlite.execute({\n sql: `INSERT INTO _test_f32blob (id, emb) VALUES (?, vector32(?))`,\n args: [\"row1\", \"[1.0, 2.0, 3.0]\"],\n });\n results[\"2_vector32_insert\"] = { success: true };\n } catch (e) {\n // Fallback: try vector() instead\n try {\n await sqlite.execute({\n sql: `INSERT INTO _test_f32blob (id, emb) VALUES (?, vector(?))`,\n args: [\"row1\", \"[1.0, 2.0, 3.0]\"],\n });\n results[\"2_vector32_insert\"] = { success: true, data: \"used vector() not vector32()\" };\n } catch (e2) {\n results[\"2_vector32_insert\"] = { success: false, error: `vector32: ${e} | vector: ${e2}` };\n }\n }\n\n // Test 3: Can we create libsql_vector_idx index?\n try {\n await sqlite.execute(`CREATE INDEX _test_f32blob_idx ON _test_f32blob (libsql_vector_idx(emb, 'metric=cosine'))`);\n results[\"3_vector_index\"] = { success: true };\n } catch (e) {\n // Fallback: try without metric\n try {\n await sqlite.execute(`CREATE INDEX _test_f32blob_idx ON _test_f32blob (libsql_vector_idx(emb))`);\n results[\"3_vector_index\"] = { success: true, data: \"no metric param\" };\n } catch (e2) {\n results[\"3_vector_index\"] = { success: false, error: `with metric: ${e} | without: ${e2}` };\n }\n }\n\n // Test 4: Insert more rows for search test\n try {\n await sqlite.execute({\n sql: `INSERT INTO _test_f32blob (id, emb) VALUES (?, vector32(?))`,\n args: [\"row2\", \"[0.0, 1.0, 0.0]\"],\n });\n await sqlite.execute({\n sql: `INSERT INTO _test_f32blob (id, emb) VALUES (?, vector32(?))`,\n args: [\"row3\", \"[0.0, 0.0, 1.0]\"],\n });\n results[\"4_multi_insert\"] = { success: true };\n } catch (e) {\n results[\"4_multi_insert\"] = { success: false, error: String(e) };\n }\n\n // Test 5: Can we use vector_top_k?\n try {\n const searchResult = await sqlite.execute({\n sql: `\n SELECT t.id \n FROM vector_top_k('_test_f32blob_idx', vector32(?), 2) v\n JOIN _test_f32blob t ON t.rowid = v.id\n `,\n args: [\"[0.9, 0.1, 0.0]\"],\n });\n results[\"5_vector_top_k\"] = { success: true, data: searchResult.rows };\n } catch (e) {\n results[\"5_vector_top_k\"] = { success: false, error: String(e) };\n }\n\n // Cleanup\n try {\n await sqlite.execute(`DROP TABLE IF EXISTS _test_f32blob`);\n } catch (_) {\n // ignore\n }\n\n // Summary\n const passed = Object.values(results).filter(r => r.success).length;\n const total = Object.keys(results).length;\n\n return Response.json({\n summary: `${passed}/${total} tests passed`,\n vector_support: passed === total,\n results,\n }, {\n headers: { \"Content-Type\": \"application/json\" },\n });\n}\n",[],[221],"files",[160,181,192,201,210,219],"counts",{"_232":33,"_233":234,"_235":33,"_236":237,"_241":242},"pulls","branches",1,"forks","envVars",{"_238":239,"_240":33},"count",4,"emptyCount","connections",{"_238":33,"_243":33},"unfilledCount","contents",[246,249,252,255,258,261],{"_105":161,"_97":162,"_163":164,"_165":-5,"_166":148,"_167":168,"_169":22,"_107":247,"_101":171,"_172":173,"_174":175,"_176":22,"_177":-5,"_100":248},["D",1769998490571],[162],{"_105":182,"_97":183,"_163":184,"_165":-5,"_166":148,"_167":185,"_169":22,"_107":250,"_101":187,"_172":188,"_174":189,"_176":22,"_177":-5,"_100":251},["D",1769996312453],[183],{"_105":193,"_97":194,"_163":195,"_165":-5,"_166":148,"_167":196,"_169":22,"_107":253,"_101":187,"_172":188,"_174":198,"_176":22,"_177":-5,"_100":254},["D",1769998333512],[194],{"_105":202,"_97":203,"_163":204,"_165":-5,"_166":148,"_167":102,"_169":22,"_107":256,"_101":33,"_172":206,"_174":207,"_176":22,"_177":-5,"_100":257},["D",1769999673928],[203],{"_105":211,"_97":212,"_163":213,"_165":-5,"_166":148,"_167":214,"_169":22,"_107":259,"_101":33,"_172":173,"_174":216,"_176":22,"_177":-5,"_100":260},["D",1769998895312],[212],{"_105":220,"_97":221,"_163":222,"_165":-5,"_166":148,"_167":223,"_169":22,"_107":262,"_101":33,"_172":206,"_174":225,"_176":22,"_177":-5,"_100":263},["D",1769999452023],[221],"upstreamProject","upstreamBranch","emptyEnvVars","isNew","flags",[],"routes/_app.x.$handle.$name.branch.$branch.version.$version.code.$",{"_95":96,"_97":98,"_99":45,"_101":102,"_100":74,"_103":272,"_158":277,"_228":296,"_188":-7,"_244":297,"_310":-5,"_311":-5,"_312":-5,"_313":-5,"_314":315,"_236":-5},{"_105":106,"_97":98,"_107":273,"_109":-5,"_110":22,"_111":112,"_113":114,"_115":-5,"_14":116,"_117":118,"_119":22,"_120":-5,"_121":-5,"_122":33,"_16":123,"_124":125,"_126":127,"_128":129,"_130":129,"_131":132,"_133":132,"_134":132,"_135":129,"_136":129,"_23":274,"_140":275,"_145":276,"_157":22},["D",1769996247298],{"_95":96,"_138":139},{"_95":96,"_142":139,"_14":116,"_143":144},{"_101":147,"_97":45,"_105":148,"_149":106,"_150":-5,"_151":-5,"_152":-5,"_153":-5,"_154":-5,"_155":98,"_156":96},[278,281,284,287,290,293],{"_105":161,"_97":162,"_163":164,"_165":-5,"_166":148,"_167":168,"_169":22,"_107":170,"_101":171,"_172":173,"_174":175,"_176":22,"_177":-5,"_178":279,"_100":280},[],[162],{"_105":182,"_97":183,"_163":184,"_165":-5,"_166":148,"_167":185,"_169":22,"_107":186,"_101":187,"_172":188,"_174":189,"_176":22,"_177":-5,"_178":282,"_100":283},[],[183],{"_105":193,"_97":194,"_163":195,"_165":-5,"_166":148,"_167":196,"_169":22,"_107":197,"_101":187,"_172":188,"_174":198,"_176":22,"_177":-5,"_178":285,"_100":286},[],[194],{"_105":202,"_97":203,"_163":204,"_165":-5,"_166":148,"_167":102,"_169":22,"_107":205,"_101":33,"_172":206,"_174":207,"_176":22,"_177":-5,"_178":288,"_100":289},[],[203],{"_105":211,"_97":212,"_163":213,"_165":-5,"_166":148,"_167":214,"_169":22,"_107":215,"_101":33,"_172":173,"_174":216,"_176":22,"_177":-5,"_178":291,"_100":292},[],[212],{"_105":220,"_97":221,"_163":222,"_165":-5,"_166":148,"_167":223,"_169":22,"_107":224,"_101":33,"_172":206,"_174":225,"_176":22,"_177":-5,"_178":294,"_100":295},[],[221],[278,281,284,287,290,293],[298,300,302,304,306,308],{"_105":161,"_97":162,"_163":164,"_165":-5,"_166":148,"_167":168,"_169":22,"_107":247,"_101":171,"_172":173,"_174":175,"_176":22,"_177":-5,"_100":299},[162],{"_105":182,"_97":183,"_163":184,"_165":-5,"_166":148,"_167":185,"_169":22,"_107":250,"_101":187,"_172":188,"_174":189,"_176":22,"_177":-5,"_100":301},[183],{"_105":193,"_97":194,"_163":195,"_165":-5,"_166":148,"_167":196,"_169":22,"_107":253,"_101":187,"_172":188,"_174":198,"_176":22,"_177":-5,"_100":303},[194],{"_105":202,"_97":203,"_163":204,"_165":-5,"_166":148,"_167":102,"_169":22,"_107":256,"_101":33,"_172":206,"_174":207,"_176":22,"_177":-5,"_100":305},[203],{"_105":211,"_97":212,"_163":213,"_165":-5,"_166":148,"_167":214,"_169":22,"_107":259,"_101":33,"_172":173,"_174":216,"_176":22,"_177":-5,"_100":307},[212],{"_105":220,"_97":221,"_163":222,"_165":-5,"_166":148,"_167":223,"_169":22,"_107":262,"_101":33,"_172":206,"_174":225,"_176":22,"_177":-5,"_100":309},[221],"upstream","pull","ahead","behind","templates",[],"actionData","errors"]

kamenxrider

slimarmor