@tekmidian/pai 0.3.2 → 0.5.0

package/dist/{postgres-CRBe30Ag.mjs.map → postgres-CIxeqf_n.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"postgres-CRBe30Ag.mjs","names":[],"sources":["../src/storage/postgres.ts"],"sourcesContent":["/**\n * PostgresBackend — implements StorageBackend using PostgreSQL + pgvector.\n *\n * Vector similarity: pgvector's <=> cosine distance operator\n * Full-text search:  PostgreSQL tsvector/tsquery (replaces SQLite FTS5)\n * Connection pooling: node-postgres Pool\n *\n * Schema is initialized via docker/init.sql.\n * This module only handles runtime queries — schema creation is external.\n */\n\nimport pg from \"pg\";\nimport type { Pool, PoolClient } from \"pg\";\nimport type { StorageBackend, ChunkRow, FileRow, FederationStats } from \"./interface.js\";\nimport type { SearchResult, SearchOptions } from \"../memory/search.js\";\nimport { buildFtsQuery } from \"../memory/search.js\";\n\nconst { Pool: PgPool } = pg;\n\n// ---------------------------------------------------------------------------\n// Postgres config\n// ---------------------------------------------------------------------------\n\nexport interface PostgresConfig {\n  connectionString?: string;\n  host?: string;\n  port?: number;\n  database?: string;\n  user?: string;\n  password?: string;\n  /** Maximum pool connections. Default 5 */\n  maxConnections?: number;\n  /** Connection timeout in ms. Default 5000 */\n  connectionTimeoutMs?: number;\n}\n\n// ---------------------------------------------------------------------------\n// Implementation\n// ---------------------------------------------------------------------------\n\nexport class PostgresBackend implements StorageBackend {\n  readonly backendType = \"postgres\" as const;\n\n  private pool: Pool;\n\n  constructor(config: PostgresConfig) {\n    const connStr =\n      config.connectionString ??\n      `postgresql://${config.user ?? \"pai\"}:${config.password ?? \"pai\"}@${config.host ?? \"localhost\"}:${config.port ?? 5432}/${config.database ?? \"pai\"}`;\n\n    this.pool = new PgPool({\n      connectionString: connStr,\n      max: config.maxConnections ?? 5,\n      connectionTimeoutMillis: config.connectionTimeoutMs ?? 5000,\n      idleTimeoutMillis: 30_000,\n    });\n\n    // Log pool errors so they don't crash the process silently\n    this.pool.on(\"error\", (err) => {\n      process.stderr.write(`[pai-postgres] Pool error: ${err.message}\\n`);\n    });\n  }\n\n  // -------------------------------------------------------------------------\n  // Lifecycle\n  // -------------------------------------------------------------------------\n\n  async close(): Promise<void> {\n    await this.pool.end();\n  }\n\n  async getStats(): Promise<FederationStats> {\n    const client = await this.pool.connect();\n    try {\n      const filesResult = await client.query<{ n: string }>(\n        \"SELECT COUNT(*)::text AS n FROM pai_files\"\n      );\n      const chunksResult = await client.query<{ n: string }>(\n        \"SELECT COUNT(*)::text AS n FROM pai_chunks\"\n      );\n      return {\n        files: parseInt(filesResult.rows[0]?.n ?? \"0\", 10),\n        chunks: parseInt(chunksResult.rows[0]?.n ?? \"0\", 10),\n      };\n    } finally {\n      client.release();\n    }\n  }\n\n  /**\n   * Test the connection by running a trivial query.\n   * Returns null on success, error message on failure.\n   */\n  async testConnection(): Promise<string | null> {\n    let client: PoolClient | null = null;\n    try {\n      client = await this.pool.connect();\n      await client.query(\"SELECT 1\");\n      return null;\n    } catch (e) {\n      return e instanceof Error ? e.message : String(e);\n    } finally {\n      client?.release();\n    }\n  }\n\n  // -------------------------------------------------------------------------\n  // File tracking\n  // -------------------------------------------------------------------------\n\n  async getFileHash(projectId: number, path: string): Promise<string | undefined> {\n    const result = await this.pool.query<{ hash: string }>(\n      \"SELECT hash FROM pai_files WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n    return result.rows[0]?.hash;\n  }\n\n  async upsertFile(file: FileRow): Promise<void> {\n    await this.pool.query(\n      `INSERT INTO pai_files (project_id, path, source, tier, hash, mtime, size)\n       VALUES ($1, $2, $3, $4, $5, $6, $7)\n       ON CONFLICT (project_id, path) DO UPDATE SET\n         source = EXCLUDED.source,\n         tier   = EXCLUDED.tier,\n         hash   = EXCLUDED.hash,\n         mtime  = EXCLUDED.mtime,\n         size   = EXCLUDED.size`,\n      [file.projectId, file.path, file.source, file.tier, file.hash, file.mtime, file.size]\n    );\n  }\n\n  // -------------------------------------------------------------------------\n  // Chunk management\n  // -------------------------------------------------------------------------\n\n  async getChunkIds(projectId: number, path: string): Promise<string[]> {\n    const result = await this.pool.query<{ id: string }>(\n      \"SELECT id FROM pai_chunks WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n    return result.rows.map((r) => r.id);\n  }\n\n  async deleteChunksForFile(projectId: number, path: string): Promise<void> {\n    // Foreign key CASCADE handles pai_chunks deletion automatically\n    // but we don't have FK to pai_chunks from pai_files, so delete explicitly\n    await this.pool.query(\n      \"DELETE FROM pai_chunks WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n  }\n\n  async insertChunks(chunks: ChunkRow[]): Promise<void> {\n    if (chunks.length === 0) return;\n\n    const client = await this.pool.connect();\n    try {\n      await client.query(\"BEGIN\");\n\n      for (const c of chunks) {\n        // embedding is null at insert time; updated separately via updateEmbedding()\n        await client.query(\n          `INSERT INTO pai_chunks\n             (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at, fts_vector)\n           VALUES\n             ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10,\n              to_tsvector('simple', $9))\n           ON CONFLICT (id) DO UPDATE SET\n             project_id = EXCLUDED.project_id,\n             source     = EXCLUDED.source,\n             tier       = EXCLUDED.tier,\n             path       = EXCLUDED.path,\n             start_line = EXCLUDED.start_line,\n             end_line   = EXCLUDED.end_line,\n             hash       = EXCLUDED.hash,\n             text       = EXCLUDED.text,\n             updated_at = EXCLUDED.updated_at,\n             fts_vector = EXCLUDED.fts_vector`,\n          [\n            c.id,\n            c.projectId,\n            c.source,\n            c.tier,\n            c.path,\n            c.startLine,\n            c.endLine,\n            c.hash,\n            c.text,\n            c.updatedAt,\n          ]\n        );\n      }\n\n      await client.query(\"COMMIT\");\n    } catch (e) {\n      await client.query(\"ROLLBACK\");\n      throw e;\n    } finally {\n      client.release();\n    }\n  }\n\n  async getUnembeddedChunkIds(projectId?: number): Promise<Array<{ id: string; text: string }>> {\n    if (projectId !== undefined) {\n      const result = await this.pool.query<{ id: string; text: string }>(\n        \"SELECT id, text FROM pai_chunks WHERE embedding IS NULL AND project_id = $1 ORDER BY id\",\n        [projectId]\n      );\n      return result.rows;\n    }\n    const result = await this.pool.query<{ id: string; text: string }>(\n      \"SELECT id, text FROM pai_chunks WHERE embedding IS NULL ORDER BY id\"\n    );\n    return result.rows;\n  }\n\n  async updateEmbedding(chunkId: string, embedding: Buffer): Promise<void> {\n    // Deserialize the Buffer (Float32Array LE bytes) to a number[] for pgvector\n    const vec = bufferToVector(embedding);\n    const vecStr = \"[\" + vec.join(\",\") + \"]\";\n    await this.pool.query(\n      \"UPDATE pai_chunks SET embedding = $1::vector WHERE id = $2\",\n      [vecStr, chunkId]\n    );\n  }\n\n  // -------------------------------------------------------------------------\n  // Search — keyword (tsvector/tsquery)\n  // -------------------------------------------------------------------------\n\n  async searchKeyword(query: string, opts?: SearchOptions): Promise<SearchResult[]> {\n    const maxResults = opts?.maxResults ?? 10;\n\n    // Build tsquery from the same token logic as buildFtsQuery, but for Postgres\n    const tsQuery = buildPgTsQuery(query);\n    if (!tsQuery) return [];\n\n    // Use 'simple' dictionary: preserves tokens as-is, no language-specific\n    // stemming. Works reliably with any language (German, French, etc.).\n    const conditions: string[] = [\"fts_vector @@ to_tsquery('simple', $1)\"];\n    const params: (string | number)[] = [tsQuery];\n    let paramIdx = 2;\n\n    if (opts?.projectIds && opts.projectIds.length > 0) {\n      const placeholders = opts.projectIds.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`project_id IN (${placeholders})`);\n      params.push(...opts.projectIds);\n    }\n\n    if (opts?.sources && opts.sources.length > 0) {\n      const placeholders = opts.sources.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`source IN (${placeholders})`);\n      params.push(...opts.sources);\n    }\n\n    if (opts?.tiers && opts.tiers.length > 0) {\n      const placeholders = opts.tiers.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`tier IN (${placeholders})`);\n      params.push(...opts.tiers);\n    }\n\n    params.push(maxResults);\n    const limitParam = `$${paramIdx}`;\n\n    const sql = `\n      SELECT\n        project_id,\n        path,\n        start_line,\n        end_line,\n        text AS snippet,\n        tier,\n        source,\n        ts_rank(fts_vector, to_tsquery('simple', $1)) AS rank_score\n      FROM pai_chunks\n      WHERE ${conditions.join(\" AND \")}\n      ORDER BY rank_score DESC\n      LIMIT ${limitParam}\n    `;\n\n    try {\n      const result = await this.pool.query<{\n        project_id: number;\n        path: string;\n        start_line: number;\n        end_line: number;\n        snippet: string;\n        tier: string;\n        source: string;\n        rank_score: number;\n      }>(sql, params);\n\n      return result.rows.map((row) => ({\n        projectId: row.project_id,\n        path: row.path,\n        startLine: row.start_line,\n        endLine: row.end_line,\n        snippet: row.snippet,\n        score: row.rank_score,\n        tier: row.tier,\n        source: row.source,\n      }));\n    } catch (e) {\n      process.stderr.write(`[pai-postgres] searchKeyword error: ${e}\\n`);\n      return [];\n    }\n  }\n\n  // -------------------------------------------------------------------------\n  // Search — semantic (pgvector cosine distance)\n  // -------------------------------------------------------------------------\n\n  async searchSemantic(queryEmbedding: Float32Array, opts?: SearchOptions): Promise<SearchResult[]> {\n    const maxResults = opts?.maxResults ?? 10;\n\n    const conditions: string[] = [\"embedding IS NOT NULL\"];\n    const params: (string | number | string)[] = [];\n    let paramIdx = 1;\n\n    // pgvector vector literal\n    const vecStr = \"[\" + Array.from(queryEmbedding).join(\",\") + \"]\";\n    params.push(vecStr);\n    const vecParam = `$${paramIdx++}`;\n\n    if (opts?.projectIds && opts.projectIds.length > 0) {\n      const placeholders = opts.projectIds.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`project_id IN (${placeholders})`);\n      params.push(...opts.projectIds);\n    }\n\n    if (opts?.sources && opts.sources.length > 0) {\n      const placeholders = opts.sources.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`source IN (${placeholders})`);\n      params.push(...opts.sources);\n    }\n\n    if (opts?.tiers && opts.tiers.length > 0) {\n      const placeholders = opts.tiers.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`tier IN (${placeholders})`);\n      params.push(...opts.tiers);\n    }\n\n    params.push(maxResults);\n    const limitParam = `$${paramIdx}`;\n\n    // <=> is cosine distance; 1 - distance = cosine similarity\n    const sql = `\n      SELECT\n        project_id,\n        path,\n        start_line,\n        end_line,\n        text AS snippet,\n        tier,\n        source,\n        1 - (embedding <=> ${vecParam}::vector) AS cosine_similarity\n      FROM pai_chunks\n      WHERE ${conditions.join(\" AND \")}\n      ORDER BY embedding <=> ${vecParam}::vector\n      LIMIT ${limitParam}\n    `;\n\n    try {\n      const result = await this.pool.query<{\n        project_id: number;\n        path: string;\n        start_line: number;\n        end_line: number;\n        snippet: string;\n        tier: string;\n        source: string;\n        cosine_similarity: number;\n      }>(sql, params);\n\n      const minScore = opts?.minScore ?? -Infinity;\n\n      return result.rows\n        .map((row) => ({\n          projectId: row.project_id,\n          path: row.path,\n          startLine: row.start_line,\n          endLine: row.end_line,\n          snippet: row.snippet,\n          score: row.cosine_similarity,\n          tier: row.tier,\n          source: row.source,\n        }))\n        .filter((r) => r.score >= minScore);\n    } catch (e) {\n      process.stderr.write(`[pai-postgres] searchSemantic error: ${e}\\n`);\n      return [];\n    }\n  }\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\n/**\n * Convert a Buffer of Float32 LE bytes (as stored in SQLite) to number[].\n */\nfunction bufferToVector(buf: Buffer): number[] {\n  const floats: number[] = [];\n  for (let i = 0; i < buf.length; i += 4) {\n    floats.push(buf.readFloatLE(i));\n  }\n  return floats;\n}\n\n/**\n * Convert a free-text query to a Postgres tsquery string.\n *\n * Uses OR (|) semantics so that a chunk matching ANY query term is returned,\n * ranked by ts_rank (which scores higher when more terms match).  AND (&)\n * semantics are too strict for multi-word queries because all terms rarely\n * co-occur in a single chunk.\n *\n * Example: \"Synchrotech interview follow-up Gilles\"\n *   → \"synchrotech | interview | follow | gilles\"\n *   → returns chunks containing any of these words, highest-matching first\n */\nfunction buildPgTsQuery(query: string): string {\n  const STOP_WORDS = new Set([\n    \"a\", \"an\", \"and\", \"are\", \"as\", \"at\", \"be\", \"been\", \"but\", \"by\",\n    \"do\", \"for\", \"from\", \"has\", \"have\", \"he\", \"her\", \"him\", \"his\",\n    \"how\", \"i\", \"if\", \"in\", \"is\", \"it\", \"its\", \"me\", \"my\", \"not\",\n    \"of\", \"on\", \"or\", \"our\", \"out\", \"she\", \"so\", \"that\", \"the\",\n    \"their\", \"them\", \"they\", \"this\", \"to\", \"up\", \"us\", \"was\", \"we\",\n    \"were\", \"what\", \"when\", \"who\", \"will\", \"with\", \"you\", \"your\",\n  ]);\n\n  const tokens = query\n    .toLowerCase()\n    .split(/[\\s\\p{P}]+/u)\n    .filter(Boolean)\n    .filter((t) => t.length >= 2)\n    .filter((t) => !STOP_WORDS.has(t))\n    // Sanitize: strip tsquery special characters to prevent syntax errors\n    .map((t) => t.replace(/'/g, \"''\").replace(/[&|!():]/g, \"\"))\n    .filter(Boolean);\n\n  if (tokens.length === 0) {\n    // Fallback: sanitize the raw query and use it as a single term\n    const raw = query.replace(/[^a-z0-9]/gi, \" \").trim().split(/\\s+/).filter(Boolean).join(\" | \");\n    return raw || \"\";\n  }\n\n  // Use OR (|) so that chunks matching ANY term are returned.\n  // ts_rank naturally scores chunks higher when more terms match, so the\n  // most relevant results still bubble to the top.\n  return tokens.join(\" | \");\n}\n\n// Re-export buildFtsQuery so it is accessible without importing search.ts\nexport { buildPgTsQuery };\n"],"mappings":";;;;;;;;;;;;;AAiBA,MAAM,EAAE,MAAM,WAAW;AAuBzB,IAAa,kBAAb,MAAuD;CACrD,AAAS,cAAc;CAEvB,AAAQ;CAER,YAAY,QAAwB;AAKlC,OAAK,OAAO,IAAI,OAAO;GACrB,kBAJA,OAAO,oBACP,gBAAgB,OAAO,QAAQ,MAAM,GAAG,OAAO,YAAY,MAAM,GAAG,OAAO,QAAQ,YAAY,GAAG,OAAO,QAAQ,KAAK,GAAG,OAAO,YAAY;GAI5I,KAAK,OAAO,kBAAkB;GAC9B,yBAAyB,OAAO,uBAAuB;GACvD,mBAAmB;GACpB,CAAC;AAGF,OAAK,KAAK,GAAG,UAAU,QAAQ;AAC7B,WAAQ,OAAO,MAAM,8BAA8B,IAAI,QAAQ,IAAI;IACnE;;CAOJ,MAAM,QAAuB;AAC3B,QAAM,KAAK,KAAK,KAAK;;CAGvB,MAAM,WAAqC;EACzC,MAAM,SAAS,MAAM,KAAK,KAAK,SAAS;AACxC,MAAI;GACF,MAAM,cAAc,MAAM,OAAO,MAC/B,4CACD;GACD,MAAM,eAAe,MAAM,OAAO,MAChC,6CACD;AACD,UAAO;IACL,OAAO,SAAS,YAAY,KAAK,IAAI,KAAK,KAAK,GAAG;IAClD,QAAQ,SAAS,aAAa,KAAK,IAAI,KAAK,KAAK,GAAG;IACrD;YACO;AACR,UAAO,SAAS;;;;;;;CAQpB,MAAM,iBAAyC;EAC7C,IAAI,SAA4B;AAChC,MAAI;AACF,YAAS,MAAM,KAAK,KAAK,SAAS;AAClC,SAAM,OAAO,MAAM,WAAW;AAC9B,UAAO;WACA,GAAG;AACV,UAAO,aAAa,QAAQ,EAAE,UAAU,OAAO,EAAE;YACzC;AACR,WAAQ,SAAS;;;CAQrB,MAAM,YAAY,WAAmB,MAA2C;AAK9E,UAJe,MAAM,KAAK,KAAK,MAC7B,kEACA,CAAC,WAAW,KAAK,CAClB,EACa,KAAK,IAAI;;CAGzB,MAAM,WAAW,MAA8B;AAC7C,QAAM,KAAK,KAAK,MACd;;;;;;;kCAQA;GAAC,KAAK;GAAW,KAAK;GAAM,KAAK;GAAQ,KAAK;GAAM,KAAK;GAAM,KAAK;GAAO,KAAK;GAAK,CACtF;;CAOH,MAAM,YAAY,WAAmB,MAAiC;AAKpE,UAJe,MAAM,KAAK,KAAK,MAC7B,iEACA,CAAC,WAAW,KAAK,CAClB,EACa,KAAK,KAAK,MAAM,EAAE,GAAG;;CAGrC,MAAM,oBAAoB,WAAmB,MAA6B;AAGxE,QAAM,KAAK,KAAK,MACd,8DACA,CAAC,WAAW,KAAK,CAClB;;CAGH,MAAM,aAAa,QAAmC;AACpD,MAAI,OAAO,WAAW,EAAG;EAEzB,MAAM,SAAS,MAAM,KAAK,KAAK,SAAS;AACxC,MAAI;AACF,SAAM,OAAO,MAAM,QAAQ;AAE3B,QAAK,MAAM,KAAK,OAEd,OAAM,OAAO,MACX;;;;;;;;;;;;;;;gDAgBA;IACE,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACH,CACF;AAGH,SAAM,OAAO,MAAM,SAAS;WACrB,GAAG;AACV,SAAM,OAAO,MAAM,WAAW;AAC9B,SAAM;YACE;AACR,UAAO,SAAS;;;CAIpB,MAAM,sBAAsB,WAAkE;AAC5F,MAAI,cAAc,OAKhB,SAJe,MAAM,KAAK,KAAK,MAC7B,2FACA,CAAC,UAAU,CACZ,EACa;AAKhB,UAHe,MAAM,KAAK,KAAK,MAC7B,sEACD,EACa;;CAGhB,MAAM,gBAAgB,SAAiB,WAAkC;EAGvE,MAAM,SAAS,MADH,eAAe,UAAU,CACZ,KAAK,IAAI,GAAG;AACrC,QAAM,KAAK,KAAK,MACd,8DACA,CAAC,QAAQ,QAAQ,CAClB;;CAOH,MAAM,cAAc,OAAe,MAA+C;EAChF,MAAM,aAAa,MAAM,cAAc;EAGvC,MAAM,UAAU,eAAe,MAAM;AACrC,MAAI,CAAC,QAAS,QAAO,EAAE;EAIvB,MAAM,aAAuB,CAAC,yCAAyC;EACvE,MAAM,SAA8B,CAAC,QAAQ;EAC7C,IAAI,WAAW;AAEf,MAAI,MAAM,cAAc,KAAK,WAAW,SAAS,GAAG;GAClD,MAAM,eAAe,KAAK,WAAW,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AAC3E,cAAW,KAAK,kBAAkB,aAAa,GAAG;AAClD,UAAO,KAAK,GAAG,KAAK,WAAW;;AAGjC,MAAI,MAAM,WAAW,KAAK,QAAQ,SAAS,GAAG;GAC5C,MAAM,eAAe,KAAK,QAAQ,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACxE,cAAW,KAAK,cAAc,aAAa,GAAG;AAC9C,UAAO,KAAK,GAAG,KAAK,QAAQ;;AAG9B,MAAI,MAAM,SAAS,KAAK,MAAM,SAAS,GAAG;GACxC,MAAM,eAAe,KAAK,MAAM,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACtE,cAAW,KAAK,YAAY,aAAa,GAAG;AAC5C,UAAO,KAAK,GAAG,KAAK,MAAM;;AAG5B,SAAO,KAAK,WAAW;EACvB,MAAM,aAAa,IAAI;EAEvB,MAAM,MAAM;;;;;;;;;;;cAWF,WAAW,KAAK,QAAQ,CAAC;;cAEzB,WAAW;;AAGrB,MAAI;AAYF,WAXe,MAAM,KAAK,KAAK,MAS5B,KAAK,OAAO,EAED,KAAK,KAAK,SAAS;IAC/B,WAAW,IAAI;IACf,MAAM,IAAI;IACV,WAAW,IAAI;IACf,SAAS,IAAI;IACb,SAAS,IAAI;IACb,OAAO,IAAI;IACX,MAAM,IAAI;IACV,QAAQ,IAAI;IACb,EAAE;WACI,GAAG;AACV,WAAQ,OAAO,MAAM,uCAAuC,EAAE,IAAI;AAClE,UAAO,EAAE;;;CAQb,MAAM,eAAe,gBAA8B,MAA+C;EAChG,MAAM,aAAa,MAAM,cAAc;EAEvC,MAAM,aAAuB,CAAC,wBAAwB;EACtD,MAAM,SAAuC,EAAE;EAC/C,IAAI,WAAW;EAGf,MAAM,SAAS,MAAM,MAAM,KAAK,eAAe,CAAC,KAAK,IAAI,GAAG;AAC5D,SAAO,KAAK,OAAO;EACnB,MAAM,WAAW,IAAI;AAErB,MAAI,MAAM,cAAc,KAAK,WAAW,SAAS,GAAG;GAClD,MAAM,eAAe,KAAK,WAAW,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AAC3E,cAAW,KAAK,kBAAkB,aAAa,GAAG;AAClD,UAAO,KAAK,GAAG,KAAK,WAAW;;AAGjC,MAAI,MAAM,WAAW,KAAK,QAAQ,SAAS,GAAG;GAC5C,MAAM,eAAe,KAAK,QAAQ,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACxE,cAAW,KAAK,cAAc,aAAa,GAAG;AAC9C,UAAO,KAAK,GAAG,KAAK,QAAQ;;AAG9B,MAAI,MAAM,SAAS,KAAK,MAAM,SAAS,GAAG;GACxC,MAAM,eAAe,KAAK,MAAM,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACtE,cAAW,KAAK,YAAY,aAAa,GAAG;AAC5C,UAAO,KAAK,GAAG,KAAK,MAAM;;AAG5B,SAAO,KAAK,WAAW;EACvB,MAAM,aAAa,IAAI;EAGvB,MAAM,MAAM;;;;;;;;;6BASa,SAAS;;cAExB,WAAW,KAAK,QAAQ,CAAC;+BACR,SAAS;cAC1B,WAAW;;AAGrB,MAAI;GACF,MAAM,SAAS,MAAM,KAAK,KAAK,MAS5B,KAAK,OAAO;GAEf,MAAM,WAAW,MAAM,YAAY;AAEnC,UAAO,OAAO,KACX,KAAK,SAAS;IACb,WAAW,IAAI;IACf,MAAM,IAAI;IACV,WAAW,IAAI;IACf,SAAS,IAAI;IACb,SAAS,IAAI;IACb,OAAO,IAAI;IACX,MAAM,IAAI;IACV,QAAQ,IAAI;IACb,EAAE,CACF,QAAQ,MAAM,EAAE,SAAS,SAAS;WAC9B,GAAG;AACV,WAAQ,OAAO,MAAM,wCAAwC,EAAE,IAAI;AACnE,UAAO,EAAE;;;;;;;AAYf,SAAS,eAAe,KAAuB;CAC7C,MAAM,SAAmB,EAAE;AAC3B,MAAK,IAAI,IAAI,GAAG,IAAI,IAAI,QAAQ,KAAK,EACnC,QAAO,KAAK,IAAI,YAAY,EAAE,CAAC;AAEjC,QAAO;;;;;;;;;;;;;;AAeT,SAAS,eAAe,OAAuB;CAC7C,MAAM,aAAa,IAAI,IAAI;EACzB;EAAK;EAAM;EAAO;EAAO;EAAM;EAAM;EAAM;EAAQ;EAAO;EAC1D;EAAM;EAAO;EAAQ;EAAO;EAAQ;EAAM;EAAO;EAAO;EACxD;EAAO;EAAK;EAAM;EAAM;EAAM;EAAM;EAAO;EAAM;EAAM;EACvD;EAAM;EAAM;EAAM;EAAO;EAAO;EAAO;EAAM;EAAQ;EACrD;EAAS;EAAQ;EAAQ;EAAQ;EAAM;EAAM;EAAM;EAAO;EAC1D;EAAQ;EAAQ;EAAQ;EAAO;EAAQ;EAAQ;EAAO;EACvD,CAAC;CAEF,MAAM,SAAS,MACZ,aAAa,CACb,MAAM,cAAc,CACpB,OAAO,QAAQ,CACf,QAAQ,MAAM,EAAE,UAAU,EAAE,CAC5B,QAAQ,MAAM,CAAC,WAAW,IAAI,EAAE,CAAC,CAEjC,KAAK,MAAM,EAAE,QAAQ,MAAM,KAAK,CAAC,QAAQ,aAAa,GAAG,CAAC,CAC1D,OAAO,QAAQ;AAElB,KAAI,OAAO,WAAW,EAGpB,QADY,MAAM,QAAQ,eAAe,IAAI,CAAC,MAAM,CAAC,MAAM,MAAM,CAAC,OAAO,QAAQ,CAAC,KAAK,MAAM,IAC/E;AAMhB,QAAO,OAAO,KAAK,MAAM"}
+{"version":3,"file":"postgres-CIxeqf_n.mjs","names":[],"sources":["../src/storage/postgres.ts"],"sourcesContent":["/**\n * PostgresBackend — implements StorageBackend using PostgreSQL + pgvector.\n *\n * Vector similarity: pgvector's <=> cosine distance operator\n * Full-text search:  PostgreSQL tsvector/tsquery (replaces SQLite FTS5)\n * Connection pooling: node-postgres Pool\n *\n * Schema is initialized via docker/init.sql.\n * This module only handles runtime queries — schema creation is external.\n */\n\nimport pg from \"pg\";\nimport type { Pool, PoolClient } from \"pg\";\nimport type { StorageBackend, ChunkRow, FileRow, FederationStats } from \"./interface.js\";\nimport type { SearchResult, SearchOptions } from \"../memory/search.js\";\nimport { buildFtsQuery } from \"../memory/search.js\";\n\nconst { Pool: PgPool } = pg;\n\n// ---------------------------------------------------------------------------\n// Postgres config\n// ---------------------------------------------------------------------------\n\nexport interface PostgresConfig {\n  connectionString?: string;\n  host?: string;\n  port?: number;\n  database?: string;\n  user?: string;\n  password?: string;\n  /** Maximum pool connections. Default 5 */\n  maxConnections?: number;\n  /** Connection timeout in ms. Default 5000 */\n  connectionTimeoutMs?: number;\n}\n\n// ---------------------------------------------------------------------------\n// Implementation\n// ---------------------------------------------------------------------------\n\nexport class PostgresBackend implements StorageBackend {\n  readonly backendType = \"postgres\" as const;\n\n  private pool: Pool;\n\n  constructor(config: PostgresConfig) {\n    const connStr =\n      config.connectionString ??\n      `postgresql://${config.user ?? \"pai\"}:${config.password ?? \"pai\"}@${config.host ?? \"localhost\"}:${config.port ?? 5432}/${config.database ?? \"pai\"}`;\n\n    this.pool = new PgPool({\n      connectionString: connStr,\n      max: config.maxConnections ?? 5,\n      connectionTimeoutMillis: config.connectionTimeoutMs ?? 5000,\n      idleTimeoutMillis: 30_000,\n    });\n\n    // Log pool errors so they don't crash the process silently\n    this.pool.on(\"error\", (err) => {\n      process.stderr.write(`[pai-postgres] Pool error: ${err.message}\\n`);\n    });\n  }\n\n  // -------------------------------------------------------------------------\n  // Lifecycle\n  // -------------------------------------------------------------------------\n\n  async close(): Promise<void> {\n    await this.pool.end();\n  }\n\n  async getStats(): Promise<FederationStats> {\n    const client = await this.pool.connect();\n    try {\n      const filesResult = await client.query<{ n: string }>(\n        \"SELECT COUNT(*)::text AS n FROM pai_files\"\n      );\n      const chunksResult = await client.query<{ n: string }>(\n        \"SELECT COUNT(*)::text AS n FROM pai_chunks\"\n      );\n      return {\n        files: parseInt(filesResult.rows[0]?.n ?? \"0\", 10),\n        chunks: parseInt(chunksResult.rows[0]?.n ?? \"0\", 10),\n      };\n    } finally {\n      client.release();\n    }\n  }\n\n  /**\n   * Test the connection by running a trivial query.\n   * Returns null on success, error message on failure.\n   */\n  async testConnection(): Promise<string | null> {\n    let client: PoolClient | null = null;\n    try {\n      client = await this.pool.connect();\n      await client.query(\"SELECT 1\");\n      return null;\n    } catch (e) {\n      return e instanceof Error ? e.message : String(e);\n    } finally {\n      client?.release();\n    }\n  }\n\n  // -------------------------------------------------------------------------\n  // File tracking\n  // -------------------------------------------------------------------------\n\n  async getFileHash(projectId: number, path: string): Promise<string | undefined> {\n    const result = await this.pool.query<{ hash: string }>(\n      \"SELECT hash FROM pai_files WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n    return result.rows[0]?.hash;\n  }\n\n  async upsertFile(file: FileRow): Promise<void> {\n    await this.pool.query(\n      `INSERT INTO pai_files (project_id, path, source, tier, hash, mtime, size)\n       VALUES ($1, $2, $3, $4, $5, $6, $7)\n       ON CONFLICT (project_id, path) DO UPDATE SET\n         source = EXCLUDED.source,\n         tier   = EXCLUDED.tier,\n         hash   = EXCLUDED.hash,\n         mtime  = EXCLUDED.mtime,\n         size   = EXCLUDED.size`,\n      [file.projectId, file.path, file.source, file.tier, file.hash, file.mtime, file.size]\n    );\n  }\n\n  // -------------------------------------------------------------------------\n  // Chunk management\n  // -------------------------------------------------------------------------\n\n  async getChunkIds(projectId: number, path: string): Promise<string[]> {\n    const result = await this.pool.query<{ id: string }>(\n      \"SELECT id FROM pai_chunks WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n    return result.rows.map((r) => r.id);\n  }\n\n  async deleteChunksForFile(projectId: number, path: string): Promise<void> {\n    // Foreign key CASCADE handles pai_chunks deletion automatically\n    // but we don't have FK to pai_chunks from pai_files, so delete explicitly\n    await this.pool.query(\n      \"DELETE FROM pai_chunks WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n  }\n\n  async insertChunks(chunks: ChunkRow[]): Promise<void> {\n    if (chunks.length === 0) return;\n\n    const client = await this.pool.connect();\n    try {\n      await client.query(\"BEGIN\");\n\n      for (const c of chunks) {\n        // embedding is null at insert time; updated separately via updateEmbedding()\n        await client.query(\n          `INSERT INTO pai_chunks\n             (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at, fts_vector)\n           VALUES\n             ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10,\n              to_tsvector('simple', $9))\n           ON CONFLICT (id) DO UPDATE SET\n             project_id = EXCLUDED.project_id,\n             source     = EXCLUDED.source,\n             tier       = EXCLUDED.tier,\n             path       = EXCLUDED.path,\n             start_line = EXCLUDED.start_line,\n             end_line   = EXCLUDED.end_line,\n             hash       = EXCLUDED.hash,\n             text       = EXCLUDED.text,\n             updated_at = EXCLUDED.updated_at,\n             fts_vector = EXCLUDED.fts_vector`,\n          [\n            c.id,\n            c.projectId,\n            c.source,\n            c.tier,\n            c.path,\n            c.startLine,\n            c.endLine,\n            c.hash,\n            c.text,\n            c.updatedAt,\n          ]\n        );\n      }\n\n      await client.query(\"COMMIT\");\n    } catch (e) {\n      await client.query(\"ROLLBACK\");\n      throw e;\n    } finally {\n      client.release();\n    }\n  }\n\n  async getUnembeddedChunkIds(projectId?: number): Promise<Array<{ id: string; text: string }>> {\n    if (projectId !== undefined) {\n      const result = await this.pool.query<{ id: string; text: string }>(\n        \"SELECT id, text FROM pai_chunks WHERE embedding IS NULL AND project_id = $1 ORDER BY id\",\n        [projectId]\n      );\n      return result.rows;\n    }\n    const result = await this.pool.query<{ id: string; text: string }>(\n      \"SELECT id, text FROM pai_chunks WHERE embedding IS NULL ORDER BY id\"\n    );\n    return result.rows;\n  }\n\n  async updateEmbedding(chunkId: string, embedding: Buffer): Promise<void> {\n    // Deserialize the Buffer (Float32Array LE bytes) to a number[] for pgvector\n    const vec = bufferToVector(embedding);\n    const vecStr = \"[\" + vec.join(\",\") + \"]\";\n    await this.pool.query(\n      \"UPDATE pai_chunks SET embedding = $1::vector WHERE id = $2\",\n      [vecStr, chunkId]\n    );\n  }\n\n  // -------------------------------------------------------------------------\n  // Search — keyword (tsvector/tsquery)\n  // -------------------------------------------------------------------------\n\n  async searchKeyword(query: string, opts?: SearchOptions): Promise<SearchResult[]> {\n    const maxResults = opts?.maxResults ?? 10;\n\n    // Build tsquery from the same token logic as buildFtsQuery, but for Postgres\n    const tsQuery = buildPgTsQuery(query);\n    if (!tsQuery) return [];\n\n    // Use 'simple' dictionary: preserves tokens as-is, no language-specific\n    // stemming. Works reliably with any language (German, French, etc.).\n    const conditions: string[] = [\"fts_vector @@ to_tsquery('simple', $1)\"];\n    const params: (string | number)[] = [tsQuery];\n    let paramIdx = 2;\n\n    if (opts?.projectIds && opts.projectIds.length > 0) {\n      const placeholders = opts.projectIds.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`project_id IN (${placeholders})`);\n      params.push(...opts.projectIds);\n    }\n\n    if (opts?.sources && opts.sources.length > 0) {\n      const placeholders = opts.sources.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`source IN (${placeholders})`);\n      params.push(...opts.sources);\n    }\n\n    if (opts?.tiers && opts.tiers.length > 0) {\n      const placeholders = opts.tiers.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`tier IN (${placeholders})`);\n      params.push(...opts.tiers);\n    }\n\n    params.push(maxResults);\n    const limitParam = `$${paramIdx}`;\n\n    const sql = `\n      SELECT\n        project_id,\n        path,\n        start_line,\n        end_line,\n        text AS snippet,\n        tier,\n        source,\n        ts_rank(fts_vector, to_tsquery('simple', $1)) AS rank_score\n      FROM pai_chunks\n      WHERE ${conditions.join(\" AND \")}\n      ORDER BY rank_score DESC\n      LIMIT ${limitParam}\n    `;\n\n    try {\n      const result = await this.pool.query<{\n        project_id: number;\n        path: string;\n        start_line: number;\n        end_line: number;\n        snippet: string;\n        tier: string;\n        source: string;\n        rank_score: number;\n      }>(sql, params);\n\n      return result.rows.map((row) => ({\n        projectId: row.project_id,\n        path: row.path,\n        startLine: row.start_line,\n        endLine: row.end_line,\n        snippet: row.snippet,\n        score: row.rank_score,\n        tier: row.tier,\n        source: row.source,\n      }));\n    } catch (e) {\n      process.stderr.write(`[pai-postgres] searchKeyword error: ${e}\\n`);\n      return [];\n    }\n  }\n\n  // -------------------------------------------------------------------------\n  // Search — semantic (pgvector cosine distance)\n  // -------------------------------------------------------------------------\n\n  async searchSemantic(queryEmbedding: Float32Array, opts?: SearchOptions): Promise<SearchResult[]> {\n    const maxResults = opts?.maxResults ?? 10;\n\n    const conditions: string[] = [\"embedding IS NOT NULL\"];\n    const params: (string | number | string)[] = [];\n    let paramIdx = 1;\n\n    // pgvector vector literal\n    const vecStr = \"[\" + Array.from(queryEmbedding).join(\",\") + \"]\";\n    params.push(vecStr);\n    const vecParam = `$${paramIdx++}`;\n\n    if (opts?.projectIds && opts.projectIds.length > 0) {\n      const placeholders = opts.projectIds.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`project_id IN (${placeholders})`);\n      params.push(...opts.projectIds);\n    }\n\n    if (opts?.sources && opts.sources.length > 0) {\n      const placeholders = opts.sources.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`source IN (${placeholders})`);\n      params.push(...opts.sources);\n    }\n\n    if (opts?.tiers && opts.tiers.length > 0) {\n      const placeholders = opts.tiers.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`tier IN (${placeholders})`);\n      params.push(...opts.tiers);\n    }\n\n    params.push(maxResults);\n    const limitParam = `$${paramIdx}`;\n\n    // <=> is cosine distance; 1 - distance = cosine similarity\n    const sql = `\n      SELECT\n        project_id,\n        path,\n        start_line,\n        end_line,\n        text AS snippet,\n        tier,\n        source,\n        1 - (embedding <=> ${vecParam}::vector) AS cosine_similarity\n      FROM pai_chunks\n      WHERE ${conditions.join(\" AND \")}\n      ORDER BY embedding <=> ${vecParam}::vector\n      LIMIT ${limitParam}\n    `;\n\n    try {\n      const result = await this.pool.query<{\n        project_id: number;\n        path: string;\n        start_line: number;\n        end_line: number;\n        snippet: string;\n        tier: string;\n        source: string;\n        cosine_similarity: number;\n      }>(sql, params);\n\n      const minScore = opts?.minScore ?? -Infinity;\n\n      return result.rows\n        .map((row) => ({\n          projectId: row.project_id,\n          path: row.path,\n          startLine: row.start_line,\n          endLine: row.end_line,\n          snippet: row.snippet,\n          score: row.cosine_similarity,\n          tier: row.tier,\n          source: row.source,\n        }))\n        .filter((r) => r.score >= minScore);\n    } catch (e) {\n      process.stderr.write(`[pai-postgres] searchSemantic error: ${e}\\n`);\n      return [];\n    }\n  }\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\n/**\n * Convert a Buffer of Float32 LE bytes (as stored in SQLite) to number[].\n */\nfunction bufferToVector(buf: Buffer): number[] {\n  const floats: number[] = [];\n  for (let i = 0; i < buf.length; i += 4) {\n    floats.push(buf.readFloatLE(i));\n  }\n  return floats;\n}\n\n/**\n * Convert a free-text query to a Postgres tsquery string.\n *\n * Uses OR (|) semantics so that a chunk matching ANY query term is returned,\n * ranked by ts_rank (which scores higher when more terms match).  AND (&)\n * semantics are too strict for multi-word queries because all terms rarely\n * co-occur in a single chunk.\n *\n * Example: \"Synchrotech interview follow-up Gilles\"\n *   → \"synchrotech | interview | follow | gilles\"\n *   → returns chunks containing any of these words, highest-matching first\n */\nfunction buildPgTsQuery(query: string): string {\n  const STOP_WORDS = new Set([\n    \"a\", \"an\", \"and\", \"are\", \"as\", \"at\", \"be\", \"been\", \"but\", \"by\",\n    \"do\", \"for\", \"from\", \"has\", \"have\", \"he\", \"her\", \"him\", \"his\",\n    \"how\", \"i\", \"if\", \"in\", \"is\", \"it\", \"its\", \"me\", \"my\", \"not\",\n    \"of\", \"on\", \"or\", \"our\", \"out\", \"she\", \"so\", \"that\", \"the\",\n    \"their\", \"them\", \"they\", \"this\", \"to\", \"up\", \"us\", \"was\", \"we\",\n    \"were\", \"what\", \"when\", \"who\", \"will\", \"with\", \"you\", \"your\",\n  ]);\n\n  const tokens = query\n    .toLowerCase()\n    .split(/[\\s\\p{P}]+/u)\n    .filter(Boolean)\n    .filter((t) => t.length >= 2)\n    .filter((t) => !STOP_WORDS.has(t))\n    // Sanitize: strip tsquery special characters to prevent syntax errors\n    .map((t) => t.replace(/'/g, \"''\").replace(/[&|!():]/g, \"\"))\n    .filter(Boolean);\n\n  if (tokens.length === 0) {\n    // Fallback: sanitize the raw query and use it as a single term\n    const raw = query.replace(/[^a-z0-9]/gi, \" \").trim().split(/\\s+/).filter(Boolean).join(\" | \");\n    return raw || \"\";\n  }\n\n  // Use OR (|) so that chunks matching ANY term are returned.\n  // ts_rank naturally scores chunks higher when more terms match, so the\n  // most relevant results still bubble to the top.\n  return tokens.join(\" | \");\n}\n\n// Re-export buildFtsQuery so it is accessible without importing search.ts\nexport { buildPgTsQuery };\n"],"mappings":";;;;;;;;;;;;;AAiBA,MAAM,EAAE,MAAM,WAAW;AAuBzB,IAAa,kBAAb,MAAuD;CACrD,AAAS,cAAc;CAEvB,AAAQ;CAER,YAAY,QAAwB;AAKlC,OAAK,OAAO,IAAI,OAAO;GACrB,kBAJA,OAAO,oBACP,gBAAgB,OAAO,QAAQ,MAAM,GAAG,OAAO,YAAY,MAAM,GAAG,OAAO,QAAQ,YAAY,GAAG,OAAO,QAAQ,KAAK,GAAG,OAAO,YAAY;GAI5I,KAAK,OAAO,kBAAkB;GAC9B,yBAAyB,OAAO,uBAAuB;GACvD,mBAAmB;GACpB,CAAC;AAGF,OAAK,KAAK,GAAG,UAAU,QAAQ;AAC7B,WAAQ,OAAO,MAAM,8BAA8B,IAAI,QAAQ,IAAI;IACnE;;CAOJ,MAAM,QAAuB;AAC3B,QAAM,KAAK,KAAK,KAAK;;CAGvB,MAAM,WAAqC;EACzC,MAAM,SAAS,MAAM,KAAK,KAAK,SAAS;AACxC,MAAI;GACF,MAAM,cAAc,MAAM,OAAO,MAC/B,4CACD;GACD,MAAM,eAAe,MAAM,OAAO,MAChC,6CACD;AACD,UAAO;IACL,OAAO,SAAS,YAAY,KAAK,IAAI,KAAK,KAAK,GAAG;IAClD,QAAQ,SAAS,aAAa,KAAK,IAAI,KAAK,KAAK,GAAG;IACrD;YACO;AACR,UAAO,SAAS;;;;;;;CAQpB,MAAM,iBAAyC;EAC7C,IAAI,SAA4B;AAChC,MAAI;AACF,YAAS,MAAM,KAAK,KAAK,SAAS;AAClC,SAAM,OAAO,MAAM,WAAW;AAC9B,UAAO;WACA,GAAG;AACV,UAAO,aAAa,QAAQ,EAAE,UAAU,OAAO,EAAE;YACzC;AACR,WAAQ,SAAS;;;CAQrB,MAAM,YAAY,WAAmB,MAA2C;AAK9E,UAJe,MAAM,KAAK,KAAK,MAC7B,kEACA,CAAC,WAAW,KAAK,CAClB,EACa,KAAK,IAAI;;CAGzB,MAAM,WAAW,MAA8B;AAC7C,QAAM,KAAK,KAAK,MACd;;;;;;;kCAQA;GAAC,KAAK;GAAW,KAAK;GAAM,KAAK;GAAQ,KAAK;GAAM,KAAK;GAAM,KAAK;GAAO,KAAK;GAAK,CACtF;;CAOH,MAAM,YAAY,WAAmB,MAAiC;AAKpE,UAJe,MAAM,KAAK,KAAK,MAC7B,iEACA,CAAC,WAAW,KAAK,CAClB,EACa,KAAK,KAAK,MAAM,EAAE,GAAG;;CAGrC,MAAM,oBAAoB,WAAmB,MAA6B;AAGxE,QAAM,KAAK,KAAK,MACd,8DACA,CAAC,WAAW,KAAK,CAClB;;CAGH,MAAM,aAAa,QAAmC;AACpD,MAAI,OAAO,WAAW,EAAG;EAEzB,MAAM,SAAS,MAAM,KAAK,KAAK,SAAS;AACxC,MAAI;AACF,SAAM,OAAO,MAAM,QAAQ;AAE3B,QAAK,MAAM,KAAK,OAEd,OAAM,OAAO,MACX;;;;;;;;;;;;;;;gDAgBA;IACE,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACF,EAAE;IACH,CACF;AAGH,SAAM,OAAO,MAAM,SAAS;WACrB,GAAG;AACV,SAAM,OAAO,MAAM,WAAW;AAC9B,SAAM;YACE;AACR,UAAO,SAAS;;;CAIpB,MAAM,sBAAsB,WAAkE;AAC5F,MAAI,cAAc,OAKhB,SAJe,MAAM,KAAK,KAAK,MAC7B,2FACA,CAAC,UAAU,CACZ,EACa;AAKhB,UAHe,MAAM,KAAK,KAAK,MAC7B,sEACD,EACa;;CAGhB,MAAM,gBAAgB,SAAiB,WAAkC;EAGvE,MAAM,SAAS,MADH,eAAe,UAAU,CACZ,KAAK,IAAI,GAAG;AACrC,QAAM,KAAK,KAAK,MACd,8DACA,CAAC,QAAQ,QAAQ,CAClB;;CAOH,MAAM,cAAc,OAAe,MAA+C;EAChF,MAAM,aAAa,MAAM,cAAc;EAGvC,MAAM,UAAU,eAAe,MAAM;AACrC,MAAI,CAAC,QAAS,QAAO,EAAE;EAIvB,MAAM,aAAuB,CAAC,yCAAyC;EACvE,MAAM,SAA8B,CAAC,QAAQ;EAC7C,IAAI,WAAW;AAEf,MAAI,MAAM,cAAc,KAAK,WAAW,SAAS,GAAG;GAClD,MAAM,eAAe,KAAK,WAAW,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AAC3E,cAAW,KAAK,kBAAkB,aAAa,GAAG;AAClD,UAAO,KAAK,GAAG,KAAK,WAAW;;AAGjC,MAAI,MAAM,WAAW,KAAK,QAAQ,SAAS,GAAG;GAC5C,MAAM,eAAe,KAAK,QAAQ,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACxE,cAAW,KAAK,cAAc,aAAa,GAAG;AAC9C,UAAO,KAAK,GAAG,KAAK,QAAQ;;AAG9B,MAAI,MAAM,SAAS,KAAK,MAAM,SAAS,GAAG;GACxC,MAAM,eAAe,KAAK,MAAM,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACtE,cAAW,KAAK,YAAY,aAAa,GAAG;AAC5C,UAAO,KAAK,GAAG,KAAK,MAAM;;AAG5B,SAAO,KAAK,WAAW;EACvB,MAAM,aAAa,IAAI;EAEvB,MAAM,MAAM;;;;;;;;;;;cAWF,WAAW,KAAK,QAAQ,CAAC;;cAEzB,WAAW;;AAGrB,MAAI;AAYF,WAXe,MAAM,KAAK,KAAK,MAS5B,KAAK,OAAO,EAED,KAAK,KAAK,SAAS;IAC/B,WAAW,IAAI;IACf,MAAM,IAAI;IACV,WAAW,IAAI;IACf,SAAS,IAAI;IACb,SAAS,IAAI;IACb,OAAO,IAAI;IACX,MAAM,IAAI;IACV,QAAQ,IAAI;IACb,EAAE;WACI,GAAG;AACV,WAAQ,OAAO,MAAM,uCAAuC,EAAE,IAAI;AAClE,UAAO,EAAE;;;CAQb,MAAM,eAAe,gBAA8B,MAA+C;EAChG,MAAM,aAAa,MAAM,cAAc;EAEvC,MAAM,aAAuB,CAAC,wBAAwB;EACtD,MAAM,SAAuC,EAAE;EAC/C,IAAI,WAAW;EAGf,MAAM,SAAS,MAAM,MAAM,KAAK,eAAe,CAAC,KAAK,IAAI,GAAG;AAC5D,SAAO,KAAK,OAAO;EACnB,MAAM,WAAW,IAAI;AAErB,MAAI,MAAM,cAAc,KAAK,WAAW,SAAS,GAAG;GAClD,MAAM,eAAe,KAAK,WAAW,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AAC3E,cAAW,KAAK,kBAAkB,aAAa,GAAG;AAClD,UAAO,KAAK,GAAG,KAAK,WAAW;;AAGjC,MAAI,MAAM,WAAW,KAAK,QAAQ,SAAS,GAAG;GAC5C,MAAM,eAAe,KAAK,QAAQ,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACxE,cAAW,KAAK,cAAc,aAAa,GAAG;AAC9C,UAAO,KAAK,GAAG,KAAK,QAAQ;;AAG9B,MAAI,MAAM,SAAS,KAAK,MAAM,SAAS,GAAG;GACxC,MAAM,eAAe,KAAK,MAAM,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACtE,cAAW,KAAK,YAAY,aAAa,GAAG;AAC5C,UAAO,KAAK,GAAG,KAAK,MAAM;;AAG5B,SAAO,KAAK,WAAW;EACvB,MAAM,aAAa,IAAI;EAGvB,MAAM,MAAM;;;;;;;;;6BASa,SAAS;;cAExB,WAAW,KAAK,QAAQ,CAAC;+BACR,SAAS;cAC1B,WAAW;;AAGrB,MAAI;GACF,MAAM,SAAS,MAAM,KAAK,KAAK,MAS5B,KAAK,OAAO;GAEf,MAAM,WAAW,MAAM,YAAY;AAEnC,UAAO,OAAO,KACX,KAAK,SAAS;IACb,WAAW,IAAI;IACf,MAAM,IAAI;IACV,WAAW,IAAI;IACf,SAAS,IAAI;IACb,SAAS,IAAI;IACb,OAAO,IAAI;IACX,MAAM,IAAI;IACV,QAAQ,IAAI;IACb,EAAE,CACF,QAAQ,MAAM,EAAE,SAAS,SAAS;WAC9B,GAAG;AACV,WAAQ,OAAO,MAAM,wCAAwC,EAAE,IAAI;AACnE,UAAO,EAAE;;;;;;;AAYf,SAAS,eAAe,KAAuB;CAC7C,MAAM,SAAmB,EAAE;AAC3B,MAAK,IAAI,IAAI,GAAG,IAAI,IAAI,QAAQ,KAAK,EACnC,QAAO,KAAK,IAAI,YAAY,EAAE,CAAC;AAEjC,QAAO;;;;;;;;;;;;;;AAeT,SAAS,eAAe,OAAuB;CAC7C,MAAM,aAAa,IAAI,IAAI;EACzB;EAAK;EAAM;EAAO;EAAO;EAAM;EAAM;EAAM;EAAQ;EAAO;EAC1D;EAAM;EAAO;EAAQ;EAAO;EAAQ;EAAM;EAAO;EAAO;EACxD;EAAO;EAAK;EAAM;EAAM;EAAM;EAAM;EAAO;EAAM;EAAM;EACvD;EAAM;EAAM;EAAM;EAAO;EAAO;EAAO;EAAM;EAAQ;EACrD;EAAS;EAAQ;EAAQ;EAAQ;EAAM;EAAM;EAAM;EAAO;EAC1D;EAAQ;EAAQ;EAAQ;EAAO;EAAQ;EAAQ;EAAO;EACvD,CAAC;CAEF,MAAM,SAAS,MACZ,aAAa,CACb,MAAM,cAAc,CACpB,OAAO,QAAQ,CACf,QAAQ,MAAM,EAAE,UAAU,EAAE,CAC5B,QAAQ,MAAM,CAAC,WAAW,IAAI,EAAE,CAAC,CAEjC,KAAK,MAAM,EAAE,QAAQ,MAAM,KAAK,CAAC,QAAQ,aAAa,GAAG,CAAC,CAC1D,OAAO,QAAQ;AAElB,KAAI,OAAO,WAAW,EAGpB,QADY,MAAM,QAAQ,eAAe,IAAI,CAAC,MAAM,CAAC,MAAM,MAAM,CAAC,OAAO,QAAQ,CAAC,KAAK,MAAM,IAC/E;AAMhB,QAAO,OAAO,KAAK,MAAM"}

package/dist/reranker-D7bRAHi6.mjs

@@ -0,0 +1,71 @@
+import { t as __exportAll } from "./rolldown-runtime-95iHPtFO.mjs";
+
+//#region src/memory/reranker.ts
+var reranker_exports = /* @__PURE__ */ __exportAll({
+	configureRerankerModel: () => configureRerankerModel,
+	rerankResults: () => rerankResults
+});
+const DEFAULT_RERANKER_MODEL = "Xenova/ms-marco-MiniLM-L-6-v2";
+let _tokenizer = null;
+let _model = null;
+let _currentModel = null;
+let _loading = null;
+/**
+* Configure the reranker model.
+* Must be called before the first rerank() call if you want a non-default model.
+*/
+function configureRerankerModel(model) {
+	const resolved = model?.trim() || DEFAULT_RERANKER_MODEL;
+	if (_currentModel !== null && _currentModel !== resolved) {
+		_tokenizer = null;
+		_model = null;
+		_loading = null;
+	}
+	_currentModel = resolved;
+}
+async function ensureLoaded() {
+	if (_tokenizer && _model) return;
+	if (_loading) return _loading;
+	_loading = (async () => {
+		const model = _currentModel ?? DEFAULT_RERANKER_MODEL;
+		const { AutoTokenizer, AutoModelForSequenceClassification } = await import("@huggingface/transformers");
+		_tokenizer = await AutoTokenizer.from_pretrained(model);
+		_model = await AutoModelForSequenceClassification.from_pretrained(model, { dtype: "q8" });
+		_currentModel = model;
+	})();
+	return _loading;
+}
+/**
+* Rerank search results using a cross-encoder model.
+*
+* Takes the top `maxCandidates` results from a first-stage retriever,
+* scores each (query, snippet) pair through the cross-encoder, and
+* returns them sorted by cross-encoder relevance score.
+*
+* The original retrieval score is replaced with the cross-encoder score.
+*/
+async function rerankResults(query, results, opts) {
+	if (results.length === 0) return [];
+	const maxCandidates = opts?.maxCandidates ?? 50;
+	const topK = opts?.topK ?? results.length;
+	const candidates = results.slice(0, maxCandidates);
+	await ensureLoaded();
+	const queries = new Array(candidates.length).fill(query);
+	const documents = candidates.map((r) => r.snippet);
+	const inputs = _tokenizer(queries, {
+		text_pair: documents,
+		padding: true,
+		truncation: true
+	});
+	const scores = (await _model(inputs)).logits.tolist();
+	const scored = candidates.map((result, i) => ({
+		...result,
+		score: scores[i][0]
+	}));
+	scored.sort((a, b) => b.score - a.score);
+	return scored.slice(0, topK);
+}
+
+//#endregion
+export { rerankResults as n, reranker_exports as r, configureRerankerModel as t };
+//# sourceMappingURL=reranker-D7bRAHi6.mjs.map

package/dist/reranker-D7bRAHi6.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"reranker-D7bRAHi6.mjs","names":[],"sources":["../src/memory/reranker.ts"],"sourcesContent":["/**\n * Cross-encoder reranker for PAI memory search results.\n *\n * Uses Xenova/ms-marco-MiniLM-L-6-v2 — a 22.7M param cross-encoder trained on\n * MS MARCO passage ranking.  The q8 quantized ONNX model is ~23 MB.\n *\n * Cross-encoders score (query, document) pairs jointly, producing more accurate\n * relevance scores than bi-encoder cosine similarity alone.  The trade-off is\n * latency: cross-encoders must score each pair independently, so they are used\n * as a reranking step on top of a fast first-stage retriever (BM25 / cosine).\n *\n * The model is loaded as a lazy singleton — no startup cost until the first\n * rerank call.  Subsequent calls reuse the loaded model.\n *\n * Inspired by QMD's Qwen3-reranker step (tobi/qmd).\n */\n\nimport type { SearchResult } from \"./search.js\";\n\n// ---------------------------------------------------------------------------\n// Constants\n// ---------------------------------------------------------------------------\n\nconst DEFAULT_RERANKER_MODEL = \"Xenova/ms-marco-MiniLM-L-6-v2\";\n\n// ---------------------------------------------------------------------------\n// Lazy singleton\n// ---------------------------------------------------------------------------\n\nlet _tokenizer: any = null;\nlet _model: any = null;\nlet _currentModel: string | null = null;\nlet _loading: Promise<void> | null = null;\n\n/**\n * Configure the reranker model.\n * Must be called before the first rerank() call if you want a non-default model.\n */\nexport function configureRerankerModel(model?: string): void {\n  const resolved = model?.trim() || DEFAULT_RERANKER_MODEL;\n  if (_currentModel !== null && _currentModel !== resolved) {\n    _tokenizer = null;\n    _model = null;\n    _loading = null;\n  }\n  _currentModel = resolved;\n}\n\nasync function ensureLoaded(): Promise<void> {\n  if (_tokenizer && _model) return;\n  if (_loading) return _loading;\n\n  _loading = (async () => {\n    const model = _currentModel ?? DEFAULT_RERANKER_MODEL;\n    const {\n      AutoTokenizer,\n      AutoModelForSequenceClassification,\n    } = await import(\"@huggingface/transformers\");\n\n    _tokenizer = await AutoTokenizer.from_pretrained(model);\n    _model = await AutoModelForSequenceClassification.from_pretrained(\n      model,\n      { dtype: \"q8\" },\n    );\n    _currentModel = model;\n  })();\n\n  return _loading;\n}\n\n// ---------------------------------------------------------------------------\n// Reranking\n// ---------------------------------------------------------------------------\n\nexport interface RerankOptions {\n  /** Maximum number of results to return after reranking. */\n  topK?: number;\n  /**\n   * Maximum number of candidates to rerank.\n   * Cross-encoders are O(n) per candidate, so we cap to keep latency\n   * reasonable.  Default: 50.\n   */\n  maxCandidates?: number;\n}\n\n/**\n * Rerank search results using a cross-encoder model.\n *\n * Takes the top `maxCandidates` results from a first-stage retriever,\n * scores each (query, snippet) pair through the cross-encoder, and\n * returns them sorted by cross-encoder relevance score.\n *\n * The original retrieval score is replaced with the cross-encoder score.\n */\nexport async function rerankResults(\n  query: string,\n  results: SearchResult[],\n  opts?: RerankOptions,\n): Promise<SearchResult[]> {\n  if (results.length === 0) return [];\n\n  const maxCandidates = opts?.maxCandidates ?? 50;\n  const topK = opts?.topK ?? results.length;\n\n  // Cap candidates to rerank\n  const candidates = results.slice(0, maxCandidates);\n\n  await ensureLoaded();\n\n  // Tokenize all (query, document) pairs in a single batch\n  const queries = new Array(candidates.length).fill(query);\n  const documents = candidates.map((r) => r.snippet);\n\n  const inputs = _tokenizer!(queries, {\n    text_pair: documents,\n    padding: true,\n    truncation: true,\n  });\n\n  // Run the cross-encoder\n  const output = await _model!(inputs);\n  const logits = output.logits;\n\n  // ms-marco-MiniLM returns raw logits (not sigmoid-normalized).\n  // Higher = more relevant.\n  const scores: number[][] = logits.tolist();\n\n  // Build reranked results\n  const scored = candidates.map((result, i) => ({\n    ...result,\n    score: scores[i][0],\n  }));\n\n  // Sort by cross-encoder score descending\n  scored.sort((a, b) => b.score - a.score);\n\n  return scored.slice(0, topK);\n}\n"],"mappings":";;;;;;;AAuBA,MAAM,yBAAyB;AAM/B,IAAI,aAAkB;AACtB,IAAI,SAAc;AAClB,IAAI,gBAA+B;AACnC,IAAI,WAAiC;;;;;AAMrC,SAAgB,uBAAuB,OAAsB;CAC3D,MAAM,WAAW,OAAO,MAAM,IAAI;AAClC,KAAI,kBAAkB,QAAQ,kBAAkB,UAAU;AACxD,eAAa;AACb,WAAS;AACT,aAAW;;AAEb,iBAAgB;;AAGlB,eAAe,eAA8B;AAC3C,KAAI,cAAc,OAAQ;AAC1B,KAAI,SAAU,QAAO;AAErB,aAAY,YAAY;EACtB,MAAM,QAAQ,iBAAiB;EAC/B,MAAM,EACJ,eACA,uCACE,MAAM,OAAO;AAEjB,eAAa,MAAM,cAAc,gBAAgB,MAAM;AACvD,WAAS,MAAM,mCAAmC,gBAChD,OACA,EAAE,OAAO,MAAM,CAChB;AACD,kBAAgB;KACd;AAEJ,QAAO;;;;;;;;;;;AA2BT,eAAsB,cACpB,OACA,SACA,MACyB;AACzB,KAAI,QAAQ,WAAW,EAAG,QAAO,EAAE;CAEnC,MAAM,gBAAgB,MAAM,iBAAiB;CAC7C,MAAM,OAAO,MAAM,QAAQ,QAAQ;CAGnC,MAAM,aAAa,QAAQ,MAAM,GAAG,cAAc;AAElD,OAAM,cAAc;CAGpB,MAAM,UAAU,IAAI,MAAM,WAAW,OAAO,CAAC,KAAK,MAAM;CACxD,MAAM,YAAY,WAAW,KAAK,MAAM,EAAE,QAAQ;CAElD,MAAM,SAAS,WAAY,SAAS;EAClC,WAAW;EACX,SAAS;EACT,YAAY;EACb,CAAC;CAQF,MAAM,UALS,MAAM,OAAQ,OAAO,EACd,OAIY,QAAQ;CAG1C,MAAM,SAAS,WAAW,KAAK,QAAQ,OAAO;EAC5C,GAAG;EACH,OAAO,OAAO,GAAG;EAClB,EAAE;AAGH,QAAO,MAAM,GAAG,MAAM,EAAE,QAAQ,EAAE,MAAM;AAExC,QAAO,OAAO,MAAM,GAAG,KAAK"}

package/dist/{schemas-BY3Pjvje.mjs → schemas-BFIgGntb.mjs}

@@ -3402,4 +3402,4 @@ const meta = meta$1;
 
 //#endregion
 export { record as a, number as i, array as n, string as o, boolean as r, unknown as s, _enum as t };
-//# sourceMappingURL=schemas-BY3Pjvje.mjs.map
+//# sourceMappingURL=schemas-BFIgGntb.mjs.map