@tekmidian/pai 0.9.0 → 0.9.1

package/dist/{detector-jGBuYQJM.mjs.map → detector-CNU3zCwP.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"detector-jGBuYQJM.mjs","names":[],"sources":["../src/topics/detector.ts"],"sourcesContent":["/**\n * Topic shift detection engine.\n *\n * Accepts a context summary (recent conversation text) and determines whether\n * the conversation has drifted away from the currently-routed project.\n *\n * Algorithm:\n *   1. Run keyword memory_search against the context text (no project filter)\n *   2. Score results by project — sum of BM25 scores per project\n *   3. Compare the top-scoring project against the current project\n *   4. If a different project dominates by more than the confidence threshold,\n *      report a topic shift.\n *\n * Design decisions:\n *   - Keyword search only (no semantic) — fast, no embedding requirement\n *   - Works with or without an active daemon (direct DB access path)\n *   - Stateless: callers supply currentProject; detector has no session memory\n *   - Minimal: returns a plain result object, not MCP content arrays\n */\n\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport { searchMemory, populateSlugs } from \"../memory/search.js\";\nimport type { SearchResult } from \"../memory/search.js\";\n\n// ---------------------------------------------------------------------------\n// Types\n// ---------------------------------------------------------------------------\n\nexport interface TopicCheckParams {\n  /** Recent conversation context (a few sentences or tool call summaries) */\n  context: string;\n  /** The project slug the session is currently routed to. May be null/empty. */\n  currentProject?: string;\n  /**\n   * Minimum confidence [0,1] to declare a shift. Default: 0.6.\n   * Higher = less sensitive, fewer false positives.\n   */\n  threshold?: number;\n  /**\n   * Maximum results to draw from memory search (candidates). Default: 20.\n   * More candidates = more accurate scoring, slightly slower.\n   */\n  candidates?: number;\n}\n\nexport interface TopicCheckResult {\n  /** Whether a significant topic shift was detected. */\n  shifted: boolean;\n  /** The project slug the session is currently routed to (echoed from input). */\n  currentProject: string | null;\n  /** The project slug that best matches the context, or null if no clear match. */\n  suggestedProject: string | null;\n  /**\n   * Confidence score for the suggested project [0,1].\n   * Represents the fraction of total score mass held by the top project.\n   * 1.0 = all matching chunks belong to one project.\n   * 0.5 = two projects are equally matched.\n   */\n  confidence: number;\n  /** Number of memory chunks that contributed to scoring. */\n  chunkCount: number;\n  /** Top-3 scoring projects with their normalised scores (for debugging). */\n  topProjects: Array<{ slug: string; score: number }>;\n}\n\n// ---------------------------------------------------------------------------\n// Core algorithm\n// ---------------------------------------------------------------------------\n\n/**\n * Detect whether the provided context text best matches a different project\n * than the session's current routing.\n *\n * Works with either a raw SQLite Database or a StorageBackend.\n * For the StorageBackend path, keyword search is used.\n * For the raw Database path (legacy/direct), searchMemory() is called.\n */\nexport async function detectTopicShift(\n  registryDb: Database,\n  federation: Database | StorageBackend,\n  params: TopicCheckParams\n): Promise<TopicCheckResult> {\n  const threshold = params.threshold ?? 0.6;\n  const candidates = params.candidates ?? 20;\n  const currentProject = params.currentProject?.trim() || null;\n\n  if (!params.context || params.context.trim().length === 0) {\n    return {\n      shifted: false,\n      currentProject,\n      suggestedProject: null,\n      confidence: 0,\n      chunkCount: 0,\n      topProjects: [],\n    };\n  }\n\n  // -------------------------------------------------------------------------\n  // Run memory search across ALL projects (no project filter)\n  // -------------------------------------------------------------------------\n\n  let results: SearchResult[];\n\n  const isBackend = (x: Database | StorageBackend): x is StorageBackend =>\n    \"backendType\" in x;\n\n  if (isBackend(federation)) {\n    results = await federation.searchKeyword(params.context, {\n      maxResults: candidates,\n    });\n  } else {\n    results = searchMemory(federation, params.context, {\n      maxResults: candidates,\n    });\n  }\n\n  if (results.length === 0) {\n    return {\n      shifted: false,\n      currentProject,\n      suggestedProject: null,\n      confidence: 0,\n      chunkCount: 0,\n      topProjects: [],\n    };\n  }\n\n  // Populate project slugs from the registry\n  const withSlugs = populateSlugs(results, registryDb);\n\n  // -------------------------------------------------------------------------\n  // Score projects by summing BM25 scores of matching chunks\n  // -------------------------------------------------------------------------\n\n  const projectScores = new Map<string, number>();\n\n  for (const r of withSlugs) {\n    const slug = r.projectSlug;\n    if (!slug) continue;\n    projectScores.set(slug, (projectScores.get(slug) ?? 0) + r.score);\n  }\n\n  if (projectScores.size === 0) {\n    return {\n      shifted: false,\n      currentProject,\n      suggestedProject: null,\n      confidence: 0,\n      chunkCount: withSlugs.length,\n      topProjects: [],\n    };\n  }\n\n  // Sort by total score descending\n  const ranked = Array.from(projectScores.entries())\n    .sort((a, b) => b[1] - a[1]);\n\n  const totalScore = ranked.reduce((sum, [, s]) => sum + s, 0);\n\n  // Top-3 for reporting (normalised to [0,1] fraction of total mass)\n  const topProjects = ranked.slice(0, 3).map(([slug, score]) => ({\n    slug,\n    score: totalScore > 0 ? score / totalScore : 0,\n  }));\n\n  const topSlug = ranked[0][0];\n  const topRawScore = ranked[0][1];\n  const confidence = totalScore > 0 ? topRawScore / totalScore : 0;\n\n  // -------------------------------------------------------------------------\n  // Determine if a shift occurred\n  // -------------------------------------------------------------------------\n\n  // A shift is detected when:\n  //   1. confidence >= threshold (the top project dominates)\n  //   2. The top project is different from currentProject\n  //   3. There is a currentProject to compare against\n  //      (if no current project, we still return the best match but no \"shift\")\n\n  const isDifferent =\n    currentProject !== null &&\n    topSlug !== currentProject;\n\n  const shifted = isDifferent && confidence >= threshold;\n\n  return {\n    shifted,\n    currentProject,\n    suggestedProject: topSlug,\n    confidence,\n    chunkCount: withSlugs.length,\n    topProjects,\n  };\n}\n"],"mappings":";;;;;;;;;;;;;AA8EA,eAAsB,iBACpB,YACA,YACA,QAC2B;CAC3B,MAAM,YAAY,OAAO,aAAa;CACtC,MAAM,aAAa,OAAO,cAAc;CACxC,MAAM,iBAAiB,OAAO,gBAAgB,MAAM,IAAI;AAExD,KAAI,CAAC,OAAO,WAAW,OAAO,QAAQ,MAAM,CAAC,WAAW,EACtD,QAAO;EACL,SAAS;EACT;EACA,kBAAkB;EAClB,YAAY;EACZ,YAAY;EACZ,aAAa,EAAE;EAChB;CAOH,IAAI;CAEJ,MAAM,aAAa,MACjB,iBAAiB;AAEnB,KAAI,UAAU,WAAW,CACvB,WAAU,MAAM,WAAW,cAAc,OAAO,SAAS,EACvD,YAAY,YACb,CAAC;KAEF,WAAU,aAAa,YAAY,OAAO,SAAS,EACjD,YAAY,YACb,CAAC;AAGJ,KAAI,QAAQ,WAAW,EACrB,QAAO;EACL,SAAS;EACT;EACA,kBAAkB;EAClB,YAAY;EACZ,YAAY;EACZ,aAAa,EAAE;EAChB;CAIH,MAAM,YAAY,cAAc,SAAS,WAAW;CAMpD,MAAM,gCAAgB,IAAI,KAAqB;AAE/C,MAAK,MAAM,KAAK,WAAW;EACzB,MAAM,OAAO,EAAE;AACf,MAAI,CAAC,KAAM;AACX,gBAAc,IAAI,OAAO,cAAc,IAAI,KAAK,IAAI,KAAK,EAAE,MAAM;;AAGnE,KAAI,cAAc,SAAS,EACzB,QAAO;EACL,SAAS;EACT;EACA,kBAAkB;EAClB,YAAY;EACZ,YAAY,UAAU;EACtB,aAAa,EAAE;EAChB;CAIH,MAAM,SAAS,MAAM,KAAK,cAAc,SAAS,CAAC,CAC/C,MAAM,GAAG,MAAM,EAAE,KAAK,EAAE,GAAG;CAE9B,MAAM,aAAa,OAAO,QAAQ,KAAK,GAAG,OAAO,MAAM,GAAG,EAAE;CAG5D,MAAM,cAAc,OAAO,MAAM,GAAG,EAAE,CAAC,KAAK,CAAC,MAAM,YAAY;EAC7D;EACA,OAAO,aAAa,IAAI,QAAQ,aAAa;EAC9C,EAAE;CAEH,MAAM,UAAU,OAAO,GAAG;CAC1B,MAAM,cAAc,OAAO,GAAG;CAC9B,MAAM,aAAa,aAAa,IAAI,cAAc,aAAa;AAkB/D,QAAO;EACL,SANA,mBAAmB,QACnB,YAAY,kBAEiB,cAAc;EAI3C;EACA,kBAAkB;EAClB;EACA,YAAY,UAAU;EACtB;EACD"}
+{"version":3,"file":"detector-CNU3zCwP.mjs","names":[],"sources":["../src/topics/detector.ts"],"sourcesContent":["/**\n * Topic shift detection engine.\n *\n * Accepts a context summary (recent conversation text) and determines whether\n * the conversation has drifted away from the currently-routed project.\n *\n * Algorithm:\n *   1. Run keyword memory_search against the context text (no project filter)\n *   2. Score results by project — sum of BM25 scores per project\n *   3. Compare the top-scoring project against the current project\n *   4. If a different project dominates by more than the confidence threshold,\n *      report a topic shift.\n *\n * Design decisions:\n *   - Keyword search only (no semantic) — fast, no embedding requirement\n *   - Works with or without an active daemon (direct DB access path)\n *   - Stateless: callers supply currentProject; detector has no session memory\n *   - Minimal: returns a plain result object, not MCP content arrays\n */\n\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport { searchMemory, populateSlugs } from \"../memory/search.js\";\nimport type { SearchResult } from \"../memory/search.js\";\n\n// ---------------------------------------------------------------------------\n// Types\n// ---------------------------------------------------------------------------\n\nexport interface TopicCheckParams {\n  /** Recent conversation context (a few sentences or tool call summaries) */\n  context: string;\n  /** The project slug the session is currently routed to. May be null/empty. */\n  currentProject?: string;\n  /**\n   * Minimum confidence [0,1] to declare a shift. Default: 0.6.\n   * Higher = less sensitive, fewer false positives.\n   */\n  threshold?: number;\n  /**\n   * Maximum results to draw from memory search (candidates). Default: 20.\n   * More candidates = more accurate scoring, slightly slower.\n   */\n  candidates?: number;\n}\n\nexport interface TopicCheckResult {\n  /** Whether a significant topic shift was detected. */\n  shifted: boolean;\n  /** The project slug the session is currently routed to (echoed from input). */\n  currentProject: string | null;\n  /** The project slug that best matches the context, or null if no clear match. */\n  suggestedProject: string | null;\n  /**\n   * Confidence score for the suggested project [0,1].\n   * Represents the fraction of total score mass held by the top project.\n   * 1.0 = all matching chunks belong to one project.\n   * 0.5 = two projects are equally matched.\n   */\n  confidence: number;\n  /** Number of memory chunks that contributed to scoring. */\n  chunkCount: number;\n  /** Top-3 scoring projects with their normalised scores (for debugging). */\n  topProjects: Array<{ slug: string; score: number }>;\n}\n\n// ---------------------------------------------------------------------------\n// Core algorithm\n// ---------------------------------------------------------------------------\n\n/**\n * Detect whether the provided context text best matches a different project\n * than the session's current routing.\n *\n * Works with either a raw SQLite Database or a StorageBackend.\n * For the StorageBackend path, keyword search is used.\n * For the raw Database path (legacy/direct), searchMemory() is called.\n */\nexport async function detectTopicShift(\n  registryDb: Database,\n  federation: Database | StorageBackend,\n  params: TopicCheckParams\n): Promise<TopicCheckResult> {\n  const threshold = params.threshold ?? 0.6;\n  const candidates = params.candidates ?? 20;\n  const currentProject = params.currentProject?.trim() || null;\n\n  if (!params.context || params.context.trim().length === 0) {\n    return {\n      shifted: false,\n      currentProject,\n      suggestedProject: null,\n      confidence: 0,\n      chunkCount: 0,\n      topProjects: [],\n    };\n  }\n\n  // -------------------------------------------------------------------------\n  // Run memory search across ALL projects (no project filter)\n  // -------------------------------------------------------------------------\n\n  let results: SearchResult[];\n\n  const isBackend = (x: Database | StorageBackend): x is StorageBackend =>\n    \"backendType\" in x;\n\n  if (isBackend(federation)) {\n    results = await federation.searchKeyword(params.context, {\n      maxResults: candidates,\n    });\n  } else {\n    results = searchMemory(federation, params.context, {\n      maxResults: candidates,\n    });\n  }\n\n  if (results.length === 0) {\n    return {\n      shifted: false,\n      currentProject,\n      suggestedProject: null,\n      confidence: 0,\n      chunkCount: 0,\n      topProjects: [],\n    };\n  }\n\n  // Populate project slugs from the registry\n  const withSlugs = populateSlugs(results, registryDb);\n\n  // -------------------------------------------------------------------------\n  // Score projects by summing BM25 scores of matching chunks\n  // -------------------------------------------------------------------------\n\n  const projectScores = new Map<string, number>();\n\n  for (const r of withSlugs) {\n    const slug = r.projectSlug;\n    if (!slug) continue;\n    projectScores.set(slug, (projectScores.get(slug) ?? 0) + r.score);\n  }\n\n  if (projectScores.size === 0) {\n    return {\n      shifted: false,\n      currentProject,\n      suggestedProject: null,\n      confidence: 0,\n      chunkCount: withSlugs.length,\n      topProjects: [],\n    };\n  }\n\n  // Sort by total score descending\n  const ranked = Array.from(projectScores.entries())\n    .sort((a, b) => b[1] - a[1]);\n\n  const totalScore = ranked.reduce((sum, [, s]) => sum + s, 0);\n\n  // Top-3 for reporting (normalised to [0,1] fraction of total mass)\n  const topProjects = ranked.slice(0, 3).map(([slug, score]) => ({\n    slug,\n    score: totalScore > 0 ? score / totalScore : 0,\n  }));\n\n  const topSlug = ranked[0][0];\n  const topRawScore = ranked[0][1];\n  const confidence = totalScore > 0 ? topRawScore / totalScore : 0;\n\n  // -------------------------------------------------------------------------\n  // Determine if a shift occurred\n  // -------------------------------------------------------------------------\n\n  // A shift is detected when:\n  //   1. confidence >= threshold (the top project dominates)\n  //   2. The top project is different from currentProject\n  //   3. There is a currentProject to compare against\n  //      (if no current project, we still return the best match but no \"shift\")\n\n  const isDifferent =\n    currentProject !== null &&\n    topSlug !== currentProject;\n\n  const shifted = isDifferent && confidence >= threshold;\n\n  return {\n    shifted,\n    currentProject,\n    suggestedProject: topSlug,\n    confidence,\n    chunkCount: withSlugs.length,\n    topProjects,\n  };\n}\n"],"mappings":";;;;;;;;;;;;;AA8EA,eAAsB,iBACpB,YACA,YACA,QAC2B;CAC3B,MAAM,YAAY,OAAO,aAAa;CACtC,MAAM,aAAa,OAAO,cAAc;CACxC,MAAM,iBAAiB,OAAO,gBAAgB,MAAM,IAAI;AAExD,KAAI,CAAC,OAAO,WAAW,OAAO,QAAQ,MAAM,CAAC,WAAW,EACtD,QAAO;EACL,SAAS;EACT;EACA,kBAAkB;EAClB,YAAY;EACZ,YAAY;EACZ,aAAa,EAAE;EAChB;CAOH,IAAI;CAEJ,MAAM,aAAa,MACjB,iBAAiB;AAEnB,KAAI,UAAU,WAAW,CACvB,WAAU,MAAM,WAAW,cAAc,OAAO,SAAS,EACvD,YAAY,YACb,CAAC;KAEF,WAAU,aAAa,YAAY,OAAO,SAAS,EACjD,YAAY,YACb,CAAC;AAGJ,KAAI,QAAQ,WAAW,EACrB,QAAO;EACL,SAAS;EACT;EACA,kBAAkB;EAClB,YAAY;EACZ,YAAY;EACZ,aAAa,EAAE;EAChB;CAIH,MAAM,YAAY,cAAc,SAAS,WAAW;CAMpD,MAAM,gCAAgB,IAAI,KAAqB;AAE/C,MAAK,MAAM,KAAK,WAAW;EACzB,MAAM,OAAO,EAAE;AACf,MAAI,CAAC,KAAM;AACX,gBAAc,IAAI,OAAO,cAAc,IAAI,KAAK,IAAI,KAAK,EAAE,MAAM;;AAGnE,KAAI,cAAc,SAAS,EACzB,QAAO;EACL,SAAS;EACT;EACA,kBAAkB;EAClB,YAAY;EACZ,YAAY,UAAU;EACtB,aAAa,EAAE;EAChB;CAIH,MAAM,SAAS,MAAM,KAAK,cAAc,SAAS,CAAC,CAC/C,MAAM,GAAG,MAAM,EAAE,KAAK,EAAE,GAAG;CAE9B,MAAM,aAAa,OAAO,QAAQ,KAAK,GAAG,OAAO,MAAM,GAAG,EAAE;CAG5D,MAAM,cAAc,OAAO,MAAM,GAAG,EAAE,CAAC,KAAK,CAAC,MAAM,YAAY;EAC7D;EACA,OAAO,aAAa,IAAI,QAAQ,aAAa;EAC9C,EAAE;CAEH,MAAM,UAAU,OAAO,GAAG;CAC1B,MAAM,cAAc,OAAO,GAAG;CAC9B,MAAM,aAAa,aAAa,IAAI,cAAc,aAAa;AAkB/D,QAAO;EACL,SANA,mBAAmB,QACnB,YAAY,kBAEiB,cAAc;EAI3C;EACA,kBAAkB;EAClB;EACA,YAAY,UAAU;EACtB;EACD"}

package/dist/{factory-e0k1HWuc.mjs → factory-DKDPRhAN.mjs}

@@ -15,7 +15,7 @@ async function createStorageBackend(config) {
 }
 async function tryPostgres(config) {
 	try {
-		const { PostgresBackend } = await import("./postgres-DvEPooLO.mjs");
+		const { PostgresBackend } = await import("./postgres-BGERehmX.mjs");
 		const pgConfig = config.postgres ?? {};
 		await PostgresBackend.ensureDatabase(pgConfig);
 		const backend = new PostgresBackend(pgConfig);
@@ -35,10 +35,10 @@ async function tryPostgres(config) {
 }
 async function createSQLiteBackend() {
 	const { openFederation } = await import("./db-DdUperSl.mjs").then((n) => n.t);
-	const { SQLiteBackend } = await import("./sqlite-l-s9xPjY.mjs");
+	const { SQLiteBackend } = await import("./sqlite-BJrME_vg.mjs");
 	return new SQLiteBackend(openFederation());
 }
 
 //#endregion
 export { factory_exports as n, createStorageBackend as t };
-//# sourceMappingURL=factory-e0k1HWuc.mjs.map
+//# sourceMappingURL=factory-DKDPRhAN.mjs.map

package/dist/{factory-e0k1HWuc.mjs.map → factory-DKDPRhAN.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"factory-e0k1HWuc.mjs","names":[],"sources":["../src/storage/factory.ts"],"sourcesContent":["/**\n * Storage backend factory.\n *\n * Reads the daemon config and returns the appropriate StorageBackend.\n * If Postgres is configured but unavailable, falls back to SQLite with\n * a warning log — the daemon never crashes due to a missing Postgres.\n */\n\nimport type { PaiDaemonConfig } from \"../daemon/config.js\";\nimport type { StorageBackend } from \"./interface.js\";\n\n/**\n * Create and return the configured StorageBackend.\n *\n * Auto-fallback behaviour:\n *  - storageBackend = \"sqlite\"   → SQLiteBackend always\n *  - storageBackend = \"postgres\" → PostgresBackend if reachable, else SQLiteBackend\n */\nexport async function createStorageBackend(\n  config: PaiDaemonConfig\n): Promise<StorageBackend> {\n  if (config.storageBackend === \"postgres\") {\n    return await tryPostgres(config);\n  }\n\n  // Default: SQLite\n  return createSQLiteBackend();\n}\n\nasync function tryPostgres(config: PaiDaemonConfig): Promise<StorageBackend> {\n  try {\n    const { PostgresBackend } = await import(\"./postgres.js\");\n    const pgConfig = config.postgres ?? {};\n\n    // Ensure the per-user database exists and has the schema applied\n    await PostgresBackend.ensureDatabase(pgConfig);\n\n    const backend = new PostgresBackend(pgConfig);\n\n    const err = await backend.testConnection();\n    if (err) {\n      process.stderr.write(\n        `[pai-daemon] Postgres unavailable (${err}). Falling back to SQLite.\\n`\n      );\n      await backend.close();\n      return createSQLiteBackend();\n    }\n\n    process.stderr.write(\"[pai-daemon] Connected to PostgreSQL backend.\\n\");\n    return backend;\n  } catch (e) {\n    const msg = e instanceof Error ? e.message : String(e);\n    process.stderr.write(\n      `[pai-daemon] Postgres init error (${msg}). Falling back to SQLite.\\n`\n    );\n    return createSQLiteBackend();\n  }\n}\n\nasync function createSQLiteBackend(): Promise<StorageBackend> {\n  const { openFederation } = await import(\"../memory/db.js\");\n  const { SQLiteBackend } = await import(\"./sqlite.js\");\n  const db = openFederation();\n  return new SQLiteBackend(db);\n}\n"],"mappings":";;;;;;;;;;;AAkBA,eAAsB,qBACpB,QACyB;AACzB,KAAI,OAAO,mBAAmB,WAC5B,QAAO,MAAM,YAAY,OAAO;AAIlC,QAAO,qBAAqB;;AAG9B,eAAe,YAAY,QAAkD;AAC3E,KAAI;EACF,MAAM,EAAE,oBAAoB,MAAM,OAAO;EACzC,MAAM,WAAW,OAAO,YAAY,EAAE;AAGtC,QAAM,gBAAgB,eAAe,SAAS;EAE9C,MAAM,UAAU,IAAI,gBAAgB,SAAS;EAE7C,MAAM,MAAM,MAAM,QAAQ,gBAAgB;AAC1C,MAAI,KAAK;AACP,WAAQ,OAAO,MACb,sCAAsC,IAAI,8BAC3C;AACD,SAAM,QAAQ,OAAO;AACrB,UAAO,qBAAqB;;AAG9B,UAAQ,OAAO,MAAM,kDAAkD;AACvE,SAAO;UACA,GAAG;EACV,MAAM,MAAM,aAAa,QAAQ,EAAE,UAAU,OAAO,EAAE;AACtD,UAAQ,OAAO,MACb,qCAAqC,IAAI,8BAC1C;AACD,SAAO,qBAAqB;;;AAIhC,eAAe,sBAA+C;CAC5D,MAAM,EAAE,mBAAmB,MAAM,OAAO;CACxC,MAAM,EAAE,kBAAkB,MAAM,OAAO;AAEvC,QAAO,IAAI,cADA,gBAAgB,CACC"}
+{"version":3,"file":"factory-DKDPRhAN.mjs","names":[],"sources":["../src/storage/factory.ts"],"sourcesContent":["/**\n * Storage backend factory.\n *\n * Reads the daemon config and returns the appropriate StorageBackend.\n * If Postgres is configured but unavailable, falls back to SQLite with\n * a warning log — the daemon never crashes due to a missing Postgres.\n */\n\nimport type { PaiDaemonConfig } from \"../daemon/config.js\";\nimport type { StorageBackend } from \"./interface.js\";\n\n/**\n * Create and return the configured StorageBackend.\n *\n * Auto-fallback behaviour:\n *  - storageBackend = \"sqlite\"   → SQLiteBackend always\n *  - storageBackend = \"postgres\" → PostgresBackend if reachable, else SQLiteBackend\n */\nexport async function createStorageBackend(\n  config: PaiDaemonConfig\n): Promise<StorageBackend> {\n  if (config.storageBackend === \"postgres\") {\n    return await tryPostgres(config);\n  }\n\n  // Default: SQLite\n  return createSQLiteBackend();\n}\n\nasync function tryPostgres(config: PaiDaemonConfig): Promise<StorageBackend> {\n  try {\n    const { PostgresBackend } = await import(\"./postgres.js\");\n    const pgConfig = config.postgres ?? {};\n\n    // Ensure the per-user database exists and has the schema applied\n    await PostgresBackend.ensureDatabase(pgConfig);\n\n    const backend = new PostgresBackend(pgConfig);\n\n    const err = await backend.testConnection();\n    if (err) {\n      process.stderr.write(\n        `[pai-daemon] Postgres unavailable (${err}). Falling back to SQLite.\\n`\n      );\n      await backend.close();\n      return createSQLiteBackend();\n    }\n\n    process.stderr.write(\"[pai-daemon] Connected to PostgreSQL backend.\\n\");\n    return backend;\n  } catch (e) {\n    const msg = e instanceof Error ? e.message : String(e);\n    process.stderr.write(\n      `[pai-daemon] Postgres init error (${msg}). Falling back to SQLite.\\n`\n    );\n    return createSQLiteBackend();\n  }\n}\n\nasync function createSQLiteBackend(): Promise<StorageBackend> {\n  const { openFederation } = await import(\"../memory/db.js\");\n  const { SQLiteBackend } = await import(\"./sqlite.js\");\n  const db = openFederation();\n  return new SQLiteBackend(db);\n}\n"],"mappings":";;;;;;;;;;;AAkBA,eAAsB,qBACpB,QACyB;AACzB,KAAI,OAAO,mBAAmB,WAC5B,QAAO,MAAM,YAAY,OAAO;AAIlC,QAAO,qBAAqB;;AAG9B,eAAe,YAAY,QAAkD;AAC3E,KAAI;EACF,MAAM,EAAE,oBAAoB,MAAM,OAAO;EACzC,MAAM,WAAW,OAAO,YAAY,EAAE;AAGtC,QAAM,gBAAgB,eAAe,SAAS;EAE9C,MAAM,UAAU,IAAI,gBAAgB,SAAS;EAE7C,MAAM,MAAM,MAAM,QAAQ,gBAAgB;AAC1C,MAAI,KAAK;AACP,WAAQ,OAAO,MACb,sCAAsC,IAAI,8BAC3C;AACD,SAAM,QAAQ,OAAO;AACrB,UAAO,qBAAqB;;AAG9B,UAAQ,OAAO,MAAM,kDAAkD;AACvE,SAAO;UACA,GAAG;EACV,MAAM,MAAM,aAAa,QAAQ,EAAE,UAAU,OAAO,EAAE;AACtD,UAAQ,OAAO,MACb,qCAAqC,IAAI,8BAC1C;AACD,SAAO,qBAAqB;;;AAIhC,eAAe,sBAA+C;CAC5D,MAAM,EAAE,mBAAmB,MAAM,OAAO;CACxC,MAAM,EAAE,kBAAkB,MAAM,OAAO;AAEvC,QAAO,IAAI,cADA,gBAAgB,CACC"}

package/dist/{indexer-backend-jcJFsmB4.mjs → indexer-backend-CIIlrYh6.mjs}

@@ -285,4 +285,4 @@ async function indexAllWithBackend(backend, registryDb) {
 
 //#endregion
 export { embedChunksWithBackend, indexAllWithBackend };
-//# sourceMappingURL=indexer-backend-jcJFsmB4.mjs.map
+//# sourceMappingURL=indexer-backend-CIIlrYh6.mjs.map

package/dist/{indexer-backend-jcJFsmB4.mjs.map → indexer-backend-CIIlrYh6.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"indexer-backend-jcJFsmB4.mjs","names":[],"sources":["../src/memory/indexer/async.ts"],"sourcesContent":["/**\n * Backend-aware async indexer for PAI federation memory.\n *\n * Provides the same functionality as sync.ts but writes through the\n * StorageBackend interface instead of directly to better-sqlite3.\n * Used when the daemon is configured with the Postgres backend.\n *\n * The SQLite path still uses sync.ts directly (which is faster for SQLite\n * due to synchronous transactions).\n */\n\nimport { readFileSync, statSync, existsSync } from \"node:fs\";\nimport { join, relative, basename } from \"node:path\";\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow } from \"../../storage/interface.js\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport {\n  sha256File,\n  chunkId,\n  detectTier,\n  walkMdFiles,\n  walkContentFiles,\n  isPathTooBroadForContentScan,\n  parseSessionTitleChunk,\n  yieldToEventLoop,\n  INDEX_YIELD_EVERY,\n} from \"./helpers.js\";\nimport type { IndexResult } from \"./types.js\";\n\nexport type { IndexResult };\n\n// ---------------------------------------------------------------------------\n// Single-file indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file through the StorageBackend interface.\n * Returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport async function indexFileWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): Promise<boolean> {\n  const absPath = join(rootPath, relativePath);\n\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Change detection\n  const existingHash = await backend.getFileHash(projectId, relativePath);\n  if (existingHash === hash) return false;\n\n  // Delete old chunks\n  await backend.deleteChunksForFile(projectId, relativePath);\n\n  // Chunk the content\n  const rawChunks = chunkMarkdown(content);\n  const updatedAt = Date.now();\n\n  const chunks: ChunkRow[] = rawChunks.map((c, i) => ({\n    id: chunkId(projectId, relativePath, i, c.startLine, c.endLine),\n    projectId,\n    source,\n    tier,\n    path: relativePath,\n    startLine: c.startLine,\n    endLine: c.endLine,\n    hash: c.hash,\n    text: c.text,\n    updatedAt,\n    embedding: null,\n  }));\n\n  // Insert chunks + update file record\n  await backend.insertChunks(chunks);\n  await backend.upsertFile({ projectId, path: relativePath, source, tier, hash, mtime, size });\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexProjectWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  const rootMemoryMd = join(rootPath, \"MEMORY.md\");\n  if (existsSync(rootMemoryMd)) {\n    filesToIndex.push({ absPath: rootMemoryMd, rootBase: rootPath, source: \"memory\", tier: \"evergreen\" });\n  }\n\n  const memoryDir = join(rootPath, \"memory\");\n  for (const absPath of walkMdFiles(memoryDir)) {\n    const relPath = relative(rootPath, absPath);\n    const tier = detectTier(relPath);\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"memory\", tier });\n  }\n\n  const notesDir = join(rootPath, \"Notes\");\n  for (const absPath of walkMdFiles(notesDir)) {\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"notes\", tier: \"session\" });\n  }\n\n  // Synthetic session-title chunks for Notes files\n  {\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const text = parseSessionTitleChunk(fileName);\n      if (!text) continue;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      const titleChunk: ChunkRow = {\n        id, projectId, source: \"notes\", tier: \"session\",\n        path: syntheticPath, startLine: 0, endLine: 0,\n        hash, text, updatedAt, embedding: null,\n      };\n      try {\n        await backend.insertChunks([titleChunk]);\n      } catch {\n        // Skip title chunks that cause backend errors\n      }\n    }\n  }\n\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        const titleChunk: ChunkRow = {\n          id, projectId, source: \"notes\", tier: \"session\",\n          path: syntheticPath, startLine: 0, endLine: 0,\n          hash, text, updatedAt, embedding: null,\n        };\n        try {\n          await backend.insertChunks([titleChunk]);\n        } catch {\n          // Skip title chunks that cause backend errors\n        }\n      }\n    }\n\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({ absPath: claudeMemoryMd, rootBase: claudeProjectDir, source: \"memory\", tier: \"evergreen\" });\n      }\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n      for (const absPath of walkMdFiles(claudeMemoryDir)) {\n        const relPath = relative(claudeProjectDir, absPath);\n        const tier = detectTier(relPath);\n        filesToIndex.push({ absPath, rootBase: claudeProjectDir, source: \"memory\", tier });\n      }\n    }\n  }\n\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    try {\n      const changed = await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier);\n\n      if (changed) {\n        const ids = await backend.getChunkIds(projectId, relPath);\n        result.filesProcessed++;\n        result.chunksCreated += ids.length;\n      } else {\n        result.filesSkipped++;\n      }\n    } catch {\n      // Skip files that cause backend errors (e.g. null bytes in Postgres)\n      result.filesSkipped++;\n    }\n  }\n\n  // Prune stale paths\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  const dbChunkPaths = await backend.getDistinctChunkPaths(projectId);\n\n  const stalePaths: string[] = [];\n  for (const p of dbChunkPaths) {\n    const basePath = p.endsWith(\"::title\") ? p.slice(0, -\"::title\".length) : p;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(p);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    await backend.deletePaths(projectId, stalePaths);\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation via StorageBackend\n// ---------------------------------------------------------------------------\n\nconst EMBED_BATCH_SIZE = 50;\nconst EMBED_YIELD_EVERY = 1;\n\n/**\n * Generate and store embeddings for all unembedded chunks via the StorageBackend.\n *\n * Processes chunks in batches of EMBED_BATCH_SIZE, yielding to the event loop\n * every EMBED_YIELD_EVERY chunks to avoid blocking IPC calls from MCP shims.\n *\n * The optional `shouldStop` callback is checked between every batch. When it\n * returns true the embed loop exits early so the caller (e.g. the daemon\n * shutdown handler) can close the pool without racing against active queries.\n *\n * Returns the number of newly embedded chunks.\n */\nexport async function embedChunksWithBackend(\n  backend: StorageBackend,\n  shouldStop?: () => boolean,\n  projectNames?: Map<number, string>,\n): Promise<number> {\n  const { generateEmbedding, serializeEmbedding } = await import(\"../embeddings.js\");\n\n  const rows = await backend.getUnembeddedChunkIds();\n  if (rows.length === 0) return 0;\n\n  const total = rows.length;\n  let embedded = 0;\n\n  // Build a summary of what needs embedding: count chunks per project_id\n  const projectChunkCounts = new Map<number, { count: number; samplePath: string }>();\n  for (const row of rows) {\n    const entry = projectChunkCounts.get(row.project_id);\n    if (entry) {\n      entry.count++;\n    } else {\n      projectChunkCounts.set(row.project_id, { count: 1, samplePath: row.path });\n    }\n  }\n  const pName = (pid: number) => projectNames?.get(pid) ?? `project ${pid}`;\n  const projectSummary = Array.from(projectChunkCounts.entries())\n    .map(([pid, { count, samplePath }]) => `  ${pName(pid)}: ${count} chunks (e.g. ${samplePath})`)\n    .join(\"\\n\");\n  process.stderr.write(\n    `[pai-daemon] Embed pass: ${total} unembedded chunks across ${projectChunkCounts.size} project(s)\\n${projectSummary}\\n`\n  );\n\n  // Track current project for transition logging\n  let currentProjectId = -1;\n  let projectEmbedded = 0;\n\n  for (let i = 0; i < rows.length; i += EMBED_BATCH_SIZE) {\n    // Check cancellation between every batch before touching the pool again\n    if (shouldStop?.()) {\n      process.stderr.write(\n        `[pai-daemon] Embed pass cancelled after ${embedded}/${total} chunks (shutdown requested)\\n`\n      );\n      break;\n    }\n\n    const batch = rows.slice(i, i + EMBED_BATCH_SIZE);\n\n    for (let j = 0; j < batch.length; j++) {\n      const { id, text, project_id, path } = batch[j];\n\n      // Log when switching to a new project\n      if (project_id !== currentProjectId) {\n        if (currentProjectId !== -1) {\n          process.stderr.write(\n            `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n          );\n        }\n        const info = projectChunkCounts.get(project_id);\n        process.stderr.write(\n          `[pai-daemon] Embedding ${pName(project_id)} (${info?.count ?? \"?\"} chunks, starting at ${path})\\n`\n        );\n        currentProjectId = project_id;\n        projectEmbedded = 0;\n      }\n\n      // Yield to the event loop periodically to keep IPC responsive\n      if ((embedded + j) % EMBED_YIELD_EVERY === 0) {\n        await yieldToEventLoop();\n      }\n\n      const vec = await generateEmbedding(text);\n      const blob = serializeEmbedding(vec);\n      await backend.updateEmbedding(id, blob);\n      projectEmbedded++;\n    }\n\n    embedded += batch.length;\n\n    // Log progress with current file path for context\n    const lastChunk = batch[batch.length - 1];\n    process.stderr.write(\n      `[pai-daemon] Embedded ${embedded}/${total} chunks (${pName(lastChunk.project_id)}: ${lastChunk.path})\\n`\n    );\n  }\n\n  // Log final project completion\n  if (currentProjectId !== -1) {\n    process.stderr.write(\n      `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n    );\n  }\n\n  return embedded;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexAllWithBackend(\n  backend: StorageBackend,\n  registryDb: Database,\n): Promise<{ projects: number; result: IndexResult }> {\n  const projects = registryDb\n    .prepare(\"SELECT id, root_path, claude_notes_dir FROM projects WHERE status = 'active'\")\n    .all() as Array<{ id: number; root_path: string; claude_notes_dir: string | null }>;\n\n  const totals: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  for (const project of projects) {\n    await yieldToEventLoop();\n    const r = await indexProjectWithBackend(backend, project.id, project.root_path, project.claude_notes_dir);\n    totals.filesProcessed += r.filesProcessed;\n    totals.chunksCreated += r.chunksCreated;\n    totals.filesSkipped += r.filesSkipped;\n  }\n\n  return { projects: projects.length, result: totals };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;AAuCA,eAAsB,qBACpB,SACA,WACA,UACA,cACA,QACA,MACkB;CAClB,MAAM,UAAU,KAAK,UAAU,aAAa;CAE5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AACN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AAIlB,KADqB,MAAM,QAAQ,YAAY,WAAW,aAAa,KAClD,KAAM,QAAO;AAGlC,OAAM,QAAQ,oBAAoB,WAAW,aAAa;CAG1D,MAAM,YAAY,cAAc,QAAQ;CACxC,MAAM,YAAY,KAAK,KAAK;CAE5B,MAAM,SAAqB,UAAU,KAAK,GAAG,OAAO;EAClD,IAAI,QAAQ,WAAW,cAAc,GAAG,EAAE,WAAW,EAAE,QAAQ;EAC/D;EACA;EACA;EACA,MAAM;EACN,WAAW,EAAE;EACb,SAAS,EAAE;EACX,MAAM,EAAE;EACR,MAAM,EAAE;EACR;EACA,WAAW;EACZ,EAAE;AAGH,OAAM,QAAQ,aAAa,OAAO;AAClC,OAAM,QAAQ,WAAW;EAAE;EAAW,MAAM;EAAc;EAAQ;EAAM;EAAM;EAAO;EAAM,CAAC;AAE5F,QAAO;;AAOT,eAAsB,wBACpB,SACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;CAEpF,MAAM,eAA2F,EAAE;CAEnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAGvG,MAAM,YAAY,KAAK,UAAU,SAAS;AAC1C,MAAK,MAAM,WAAW,YAAY,UAAU,EAAE;EAE5C,MAAM,OAAO,WADG,SAAS,UAAU,QAAQ,CACX;AAChC,eAAa,KAAK;GAAE;GAAS,UAAU;GAAU,QAAQ;GAAU;GAAM,CAAC;;CAG5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAItF;EACE,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAE3C,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,OAAI,CAAC,KAAM;GAEX,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GAGjC,MAAM,aAAuB;IAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IAG/C;IAAW,QAAQ;IAAS,MAAM;IACtC,MAAM;IAAe,WAAW;IAAG,SAAS;IAC5C,MAJW,WAAW,KAAK;IAIrB;IAAM;IAAW,WAAW;IACnC;AACD,OAAI;AACF,UAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;WAClC;;;AAMZ,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAIxF,KAAI,kBAAkB,mBAAmB,UAAU;AACjD,OAAK,MAAM,WAAW,YAAY,eAAe,CAC/C,cAAa,KAAK;GAAE;GAAS,UAAU;GAAgB,QAAQ;GAAS,MAAM;GAAW,CAAC;EAI5F;GACE,MAAM,YAAY,KAAK,KAAK;AAC5B,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IAEjD,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,QAAI,CAAC,KAAM;IAEX,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IAGjC,MAAM,aAAuB;KAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;KAG/C;KAAW,QAAQ;KAAS,MAAM;KACtC,MAAM;KAAe,WAAW;KAAG,SAAS;KAC5C,MAJW,WAAW,KAAK;KAIrB;KAAM;KAAW,WAAW;KACnC;AACD,QAAI;AACF,WAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;YAClC;;;AAMZ,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAAE,SAAS;IAAgB,UAAU;IAAkB,QAAQ;IAAU,MAAM;IAAa,CAAC;GAEjH,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;AACxD,QAAK,MAAM,WAAW,YAAY,gBAAgB,EAAE;IAElD,MAAM,OAAO,WADG,SAAS,kBAAkB,QAAQ,CACnB;AAChC,iBAAa,KAAK;KAAE;KAAS,UAAU;KAAkB,QAAQ;KAAU;KAAM,CAAC;;;;AAKxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAC9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAC3C,MAAI;AAGF,OAFgB,MAAM,qBAAqB,SAAS,WAAW,UAAU,SAAS,QAAQ,KAAK,EAElF;IACX,MAAM,MAAM,MAAM,QAAQ,YAAY,WAAW,QAAQ;AACzD,WAAO;AACP,WAAO,iBAAiB,IAAI;SAE5B,QAAO;UAEH;AAEN,UAAO;;;CAKX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAG5C,MAAM,eAAe,MAAM,QAAQ,sBAAsB,UAAU;CAEnE,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,KAAK,cAAc;EAC5B,MAAM,WAAW,EAAE,SAAS,UAAU,GAAG,EAAE,MAAM,GAAG,GAAkB,GAAG;AACzE,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,EAAE;;AAItB,KAAI,WAAW,SAAS,EACtB,OAAM,QAAQ,YAAY,WAAW,WAAW;AAGlD,QAAO;;AAOT,MAAM,mBAAmB;AACzB,MAAM,oBAAoB;;;;;;;;;;;;;AAc1B,eAAsB,uBACpB,SACA,YACA,cACiB;CACjB,MAAM,EAAE,mBAAmB,uBAAuB,MAAM,OAAO;CAE/D,MAAM,OAAO,MAAM,QAAQ,uBAAuB;AAClD,KAAI,KAAK,WAAW,EAAG,QAAO;CAE9B,MAAM,QAAQ,KAAK;CACnB,IAAI,WAAW;CAGf,MAAM,qCAAqB,IAAI,KAAoD;AACnF,MAAK,MAAM,OAAO,MAAM;EACtB,MAAM,QAAQ,mBAAmB,IAAI,IAAI,WAAW;AACpD,MAAI,MACF,OAAM;MAEN,oBAAmB,IAAI,IAAI,YAAY;GAAE,OAAO;GAAG,YAAY,IAAI;GAAM,CAAC;;CAG9E,MAAM,SAAS,QAAgB,cAAc,IAAI,IAAI,IAAI,WAAW;CACpE,MAAM,iBAAiB,MAAM,KAAK,mBAAmB,SAAS,CAAC,CAC5D,KAAK,CAAC,KAAK,EAAE,OAAO,kBAAkB,KAAK,MAAM,IAAI,CAAC,IAAI,MAAM,gBAAgB,WAAW,GAAG,CAC9F,KAAK,KAAK;AACb,SAAQ,OAAO,MACb,4BAA4B,MAAM,4BAA4B,mBAAmB,KAAK,eAAe,eAAe,IACrH;CAGD,IAAI,mBAAmB;CACvB,IAAI,kBAAkB;AAEtB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,kBAAkB;AAEtD,MAAI,cAAc,EAAE;AAClB,WAAQ,OAAO,MACb,2CAA2C,SAAS,GAAG,MAAM,gCAC9D;AACD;;EAGF,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,iBAAiB;AAEjD,OAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;GACrC,MAAM,EAAE,IAAI,MAAM,YAAY,SAAS,MAAM;AAG7C,OAAI,eAAe,kBAAkB;AACnC,QAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;IAEH,MAAM,OAAO,mBAAmB,IAAI,WAAW;AAC/C,YAAQ,OAAO,MACb,0BAA0B,MAAM,WAAW,CAAC,IAAI,MAAM,SAAS,IAAI,uBAAuB,KAAK,KAChG;AACD,uBAAmB;AACnB,sBAAkB;;AAIpB,QAAK,WAAW,KAAK,sBAAsB,EACzC,OAAM,kBAAkB;GAI1B,MAAM,OAAO,mBADD,MAAM,kBAAkB,KAAK,CACL;AACpC,SAAM,QAAQ,gBAAgB,IAAI,KAAK;AACvC;;AAGF,cAAY,MAAM;EAGlB,MAAM,YAAY,MAAM,MAAM,SAAS;AACvC,UAAQ,OAAO,MACb,yBAAyB,SAAS,GAAG,MAAM,WAAW,MAAM,UAAU,WAAW,CAAC,IAAI,UAAU,KAAK,KACtG;;AAIH,KAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;AAGH,QAAO;;AAOT,eAAsB,oBACpB,SACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;AAEpF,MAAK,MAAM,WAAW,UAAU;AAC9B,QAAM,kBAAkB;EACxB,MAAM,IAAI,MAAM,wBAAwB,SAAS,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzG,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ"}
+{"version":3,"file":"indexer-backend-CIIlrYh6.mjs","names":[],"sources":["../src/memory/indexer/async.ts"],"sourcesContent":["/**\n * Backend-aware async indexer for PAI federation memory.\n *\n * Provides the same functionality as sync.ts but writes through the\n * StorageBackend interface instead of directly to better-sqlite3.\n * Used when the daemon is configured with the Postgres backend.\n *\n * The SQLite path still uses sync.ts directly (which is faster for SQLite\n * due to synchronous transactions).\n */\n\nimport { readFileSync, statSync, existsSync } from \"node:fs\";\nimport { join, relative, basename } from \"node:path\";\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow } from \"../../storage/interface.js\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport {\n  sha256File,\n  chunkId,\n  detectTier,\n  walkMdFiles,\n  walkContentFiles,\n  isPathTooBroadForContentScan,\n  parseSessionTitleChunk,\n  yieldToEventLoop,\n  INDEX_YIELD_EVERY,\n} from \"./helpers.js\";\nimport type { IndexResult } from \"./types.js\";\n\nexport type { IndexResult };\n\n// ---------------------------------------------------------------------------\n// Single-file indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file through the StorageBackend interface.\n * Returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport async function indexFileWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): Promise<boolean> {\n  const absPath = join(rootPath, relativePath);\n\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Change detection\n  const existingHash = await backend.getFileHash(projectId, relativePath);\n  if (existingHash === hash) return false;\n\n  // Delete old chunks\n  await backend.deleteChunksForFile(projectId, relativePath);\n\n  // Chunk the content\n  const rawChunks = chunkMarkdown(content);\n  const updatedAt = Date.now();\n\n  const chunks: ChunkRow[] = rawChunks.map((c, i) => ({\n    id: chunkId(projectId, relativePath, i, c.startLine, c.endLine),\n    projectId,\n    source,\n    tier,\n    path: relativePath,\n    startLine: c.startLine,\n    endLine: c.endLine,\n    hash: c.hash,\n    text: c.text,\n    updatedAt,\n    embedding: null,\n  }));\n\n  // Insert chunks + update file record\n  await backend.insertChunks(chunks);\n  await backend.upsertFile({ projectId, path: relativePath, source, tier, hash, mtime, size });\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexProjectWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  const rootMemoryMd = join(rootPath, \"MEMORY.md\");\n  if (existsSync(rootMemoryMd)) {\n    filesToIndex.push({ absPath: rootMemoryMd, rootBase: rootPath, source: \"memory\", tier: \"evergreen\" });\n  }\n\n  const memoryDir = join(rootPath, \"memory\");\n  for (const absPath of walkMdFiles(memoryDir)) {\n    const relPath = relative(rootPath, absPath);\n    const tier = detectTier(relPath);\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"memory\", tier });\n  }\n\n  const notesDir = join(rootPath, \"Notes\");\n  for (const absPath of walkMdFiles(notesDir)) {\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"notes\", tier: \"session\" });\n  }\n\n  // Synthetic session-title chunks for Notes files\n  {\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const text = parseSessionTitleChunk(fileName);\n      if (!text) continue;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      const titleChunk: ChunkRow = {\n        id, projectId, source: \"notes\", tier: \"session\",\n        path: syntheticPath, startLine: 0, endLine: 0,\n        hash, text, updatedAt, embedding: null,\n      };\n      try {\n        await backend.insertChunks([titleChunk]);\n      } catch {\n        // Skip title chunks that cause backend errors\n      }\n    }\n  }\n\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        const titleChunk: ChunkRow = {\n          id, projectId, source: \"notes\", tier: \"session\",\n          path: syntheticPath, startLine: 0, endLine: 0,\n          hash, text, updatedAt, embedding: null,\n        };\n        try {\n          await backend.insertChunks([titleChunk]);\n        } catch {\n          // Skip title chunks that cause backend errors\n        }\n      }\n    }\n\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({ absPath: claudeMemoryMd, rootBase: claudeProjectDir, source: \"memory\", tier: \"evergreen\" });\n      }\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n      for (const absPath of walkMdFiles(claudeMemoryDir)) {\n        const relPath = relative(claudeProjectDir, absPath);\n        const tier = detectTier(relPath);\n        filesToIndex.push({ absPath, rootBase: claudeProjectDir, source: \"memory\", tier });\n      }\n    }\n  }\n\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    try {\n      const changed = await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier);\n\n      if (changed) {\n        const ids = await backend.getChunkIds(projectId, relPath);\n        result.filesProcessed++;\n        result.chunksCreated += ids.length;\n      } else {\n        result.filesSkipped++;\n      }\n    } catch {\n      // Skip files that cause backend errors (e.g. null bytes in Postgres)\n      result.filesSkipped++;\n    }\n  }\n\n  // Prune stale paths\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  const dbChunkPaths = await backend.getDistinctChunkPaths(projectId);\n\n  const stalePaths: string[] = [];\n  for (const p of dbChunkPaths) {\n    const basePath = p.endsWith(\"::title\") ? p.slice(0, -\"::title\".length) : p;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(p);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    await backend.deletePaths(projectId, stalePaths);\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation via StorageBackend\n// ---------------------------------------------------------------------------\n\nconst EMBED_BATCH_SIZE = 50;\nconst EMBED_YIELD_EVERY = 1;\n\n/**\n * Generate and store embeddings for all unembedded chunks via the StorageBackend.\n *\n * Processes chunks in batches of EMBED_BATCH_SIZE, yielding to the event loop\n * every EMBED_YIELD_EVERY chunks to avoid blocking IPC calls from MCP shims.\n *\n * The optional `shouldStop` callback is checked between every batch. When it\n * returns true the embed loop exits early so the caller (e.g. the daemon\n * shutdown handler) can close the pool without racing against active queries.\n *\n * Returns the number of newly embedded chunks.\n */\nexport async function embedChunksWithBackend(\n  backend: StorageBackend,\n  shouldStop?: () => boolean,\n  projectNames?: Map<number, string>,\n): Promise<number> {\n  const { generateEmbedding, serializeEmbedding } = await import(\"../embeddings.js\");\n\n  const rows = await backend.getUnembeddedChunkIds();\n  if (rows.length === 0) return 0;\n\n  const total = rows.length;\n  let embedded = 0;\n\n  // Build a summary of what needs embedding: count chunks per project_id\n  const projectChunkCounts = new Map<number, { count: number; samplePath: string }>();\n  for (const row of rows) {\n    const entry = projectChunkCounts.get(row.project_id);\n    if (entry) {\n      entry.count++;\n    } else {\n      projectChunkCounts.set(row.project_id, { count: 1, samplePath: row.path });\n    }\n  }\n  const pName = (pid: number) => projectNames?.get(pid) ?? `project ${pid}`;\n  const projectSummary = Array.from(projectChunkCounts.entries())\n    .map(([pid, { count, samplePath }]) => `  ${pName(pid)}: ${count} chunks (e.g. ${samplePath})`)\n    .join(\"\\n\");\n  process.stderr.write(\n    `[pai-daemon] Embed pass: ${total} unembedded chunks across ${projectChunkCounts.size} project(s)\\n${projectSummary}\\n`\n  );\n\n  // Track current project for transition logging\n  let currentProjectId = -1;\n  let projectEmbedded = 0;\n\n  for (let i = 0; i < rows.length; i += EMBED_BATCH_SIZE) {\n    // Check cancellation between every batch before touching the pool again\n    if (shouldStop?.()) {\n      process.stderr.write(\n        `[pai-daemon] Embed pass cancelled after ${embedded}/${total} chunks (shutdown requested)\\n`\n      );\n      break;\n    }\n\n    const batch = rows.slice(i, i + EMBED_BATCH_SIZE);\n\n    for (let j = 0; j < batch.length; j++) {\n      const { id, text, project_id, path } = batch[j];\n\n      // Log when switching to a new project\n      if (project_id !== currentProjectId) {\n        if (currentProjectId !== -1) {\n          process.stderr.write(\n            `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n          );\n        }\n        const info = projectChunkCounts.get(project_id);\n        process.stderr.write(\n          `[pai-daemon] Embedding ${pName(project_id)} (${info?.count ?? \"?\"} chunks, starting at ${path})\\n`\n        );\n        currentProjectId = project_id;\n        projectEmbedded = 0;\n      }\n\n      // Yield to the event loop periodically to keep IPC responsive\n      if ((embedded + j) % EMBED_YIELD_EVERY === 0) {\n        await yieldToEventLoop();\n      }\n\n      const vec = await generateEmbedding(text);\n      const blob = serializeEmbedding(vec);\n      await backend.updateEmbedding(id, blob);\n      projectEmbedded++;\n    }\n\n    embedded += batch.length;\n\n    // Log progress with current file path for context\n    const lastChunk = batch[batch.length - 1];\n    process.stderr.write(\n      `[pai-daemon] Embedded ${embedded}/${total} chunks (${pName(lastChunk.project_id)}: ${lastChunk.path})\\n`\n    );\n  }\n\n  // Log final project completion\n  if (currentProjectId !== -1) {\n    process.stderr.write(\n      `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n    );\n  }\n\n  return embedded;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexAllWithBackend(\n  backend: StorageBackend,\n  registryDb: Database,\n): Promise<{ projects: number; result: IndexResult }> {\n  const projects = registryDb\n    .prepare(\"SELECT id, root_path, claude_notes_dir FROM projects WHERE status = 'active'\")\n    .all() as Array<{ id: number; root_path: string; claude_notes_dir: string | null }>;\n\n  const totals: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  for (const project of projects) {\n    await yieldToEventLoop();\n    const r = await indexProjectWithBackend(backend, project.id, project.root_path, project.claude_notes_dir);\n    totals.filesProcessed += r.filesProcessed;\n    totals.chunksCreated += r.chunksCreated;\n    totals.filesSkipped += r.filesSkipped;\n  }\n\n  return { projects: projects.length, result: totals };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;AAuCA,eAAsB,qBACpB,SACA,WACA,UACA,cACA,QACA,MACkB;CAClB,MAAM,UAAU,KAAK,UAAU,aAAa;CAE5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AACN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AAIlB,KADqB,MAAM,QAAQ,YAAY,WAAW,aAAa,KAClD,KAAM,QAAO;AAGlC,OAAM,QAAQ,oBAAoB,WAAW,aAAa;CAG1D,MAAM,YAAY,cAAc,QAAQ;CACxC,MAAM,YAAY,KAAK,KAAK;CAE5B,MAAM,SAAqB,UAAU,KAAK,GAAG,OAAO;EAClD,IAAI,QAAQ,WAAW,cAAc,GAAG,EAAE,WAAW,EAAE,QAAQ;EAC/D;EACA;EACA;EACA,MAAM;EACN,WAAW,EAAE;EACb,SAAS,EAAE;EACX,MAAM,EAAE;EACR,MAAM,EAAE;EACR;EACA,WAAW;EACZ,EAAE;AAGH,OAAM,QAAQ,aAAa,OAAO;AAClC,OAAM,QAAQ,WAAW;EAAE;EAAW,MAAM;EAAc;EAAQ;EAAM;EAAM;EAAO;EAAM,CAAC;AAE5F,QAAO;;AAOT,eAAsB,wBACpB,SACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;CAEpF,MAAM,eAA2F,EAAE;CAEnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAGvG,MAAM,YAAY,KAAK,UAAU,SAAS;AAC1C,MAAK,MAAM,WAAW,YAAY,UAAU,EAAE;EAE5C,MAAM,OAAO,WADG,SAAS,UAAU,QAAQ,CACX;AAChC,eAAa,KAAK;GAAE;GAAS,UAAU;GAAU,QAAQ;GAAU;GAAM,CAAC;;CAG5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAItF;EACE,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAE3C,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,OAAI,CAAC,KAAM;GAEX,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GAGjC,MAAM,aAAuB;IAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IAG/C;IAAW,QAAQ;IAAS,MAAM;IACtC,MAAM;IAAe,WAAW;IAAG,SAAS;IAC5C,MAJW,WAAW,KAAK;IAIrB;IAAM;IAAW,WAAW;IACnC;AACD,OAAI;AACF,UAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;WAClC;;;AAMZ,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAIxF,KAAI,kBAAkB,mBAAmB,UAAU;AACjD,OAAK,MAAM,WAAW,YAAY,eAAe,CAC/C,cAAa,KAAK;GAAE;GAAS,UAAU;GAAgB,QAAQ;GAAS,MAAM;GAAW,CAAC;EAI5F;GACE,MAAM,YAAY,KAAK,KAAK;AAC5B,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IAEjD,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,QAAI,CAAC,KAAM;IAEX,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IAGjC,MAAM,aAAuB;KAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;KAG/C;KAAW,QAAQ;KAAS,MAAM;KACtC,MAAM;KAAe,WAAW;KAAG,SAAS;KAC5C,MAJW,WAAW,KAAK;KAIrB;KAAM;KAAW,WAAW;KACnC;AACD,QAAI;AACF,WAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;YAClC;;;AAMZ,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAAE,SAAS;IAAgB,UAAU;IAAkB,QAAQ;IAAU,MAAM;IAAa,CAAC;GAEjH,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;AACxD,QAAK,MAAM,WAAW,YAAY,gBAAgB,EAAE;IAElD,MAAM,OAAO,WADG,SAAS,kBAAkB,QAAQ,CACnB;AAChC,iBAAa,KAAK;KAAE;KAAS,UAAU;KAAkB,QAAQ;KAAU;KAAM,CAAC;;;;AAKxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAC9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAC3C,MAAI;AAGF,OAFgB,MAAM,qBAAqB,SAAS,WAAW,UAAU,SAAS,QAAQ,KAAK,EAElF;IACX,MAAM,MAAM,MAAM,QAAQ,YAAY,WAAW,QAAQ;AACzD,WAAO;AACP,WAAO,iBAAiB,IAAI;SAE5B,QAAO;UAEH;AAEN,UAAO;;;CAKX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAG5C,MAAM,eAAe,MAAM,QAAQ,sBAAsB,UAAU;CAEnE,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,KAAK,cAAc;EAC5B,MAAM,WAAW,EAAE,SAAS,UAAU,GAAG,EAAE,MAAM,GAAG,GAAkB,GAAG;AACzE,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,EAAE;;AAItB,KAAI,WAAW,SAAS,EACtB,OAAM,QAAQ,YAAY,WAAW,WAAW;AAGlD,QAAO;;AAOT,MAAM,mBAAmB;AACzB,MAAM,oBAAoB;;;;;;;;;;;;;AAc1B,eAAsB,uBACpB,SACA,YACA,cACiB;CACjB,MAAM,EAAE,mBAAmB,uBAAuB,MAAM,OAAO;CAE/D,MAAM,OAAO,MAAM,QAAQ,uBAAuB;AAClD,KAAI,KAAK,WAAW,EAAG,QAAO;CAE9B,MAAM,QAAQ,KAAK;CACnB,IAAI,WAAW;CAGf,MAAM,qCAAqB,IAAI,KAAoD;AACnF,MAAK,MAAM,OAAO,MAAM;EACtB,MAAM,QAAQ,mBAAmB,IAAI,IAAI,WAAW;AACpD,MAAI,MACF,OAAM;MAEN,oBAAmB,IAAI,IAAI,YAAY;GAAE,OAAO;GAAG,YAAY,IAAI;GAAM,CAAC;;CAG9E,MAAM,SAAS,QAAgB,cAAc,IAAI,IAAI,IAAI,WAAW;CACpE,MAAM,iBAAiB,MAAM,KAAK,mBAAmB,SAAS,CAAC,CAC5D,KAAK,CAAC,KAAK,EAAE,OAAO,kBAAkB,KAAK,MAAM,IAAI,CAAC,IAAI,MAAM,gBAAgB,WAAW,GAAG,CAC9F,KAAK,KAAK;AACb,SAAQ,OAAO,MACb,4BAA4B,MAAM,4BAA4B,mBAAmB,KAAK,eAAe,eAAe,IACrH;CAGD,IAAI,mBAAmB;CACvB,IAAI,kBAAkB;AAEtB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,kBAAkB;AAEtD,MAAI,cAAc,EAAE;AAClB,WAAQ,OAAO,MACb,2CAA2C,SAAS,GAAG,MAAM,gCAC9D;AACD;;EAGF,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,iBAAiB;AAEjD,OAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;GACrC,MAAM,EAAE,IAAI,MAAM,YAAY,SAAS,MAAM;AAG7C,OAAI,eAAe,kBAAkB;AACnC,QAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;IAEH,MAAM,OAAO,mBAAmB,IAAI,WAAW;AAC/C,YAAQ,OAAO,MACb,0BAA0B,MAAM,WAAW,CAAC,IAAI,MAAM,SAAS,IAAI,uBAAuB,KAAK,KAChG;AACD,uBAAmB;AACnB,sBAAkB;;AAIpB,QAAK,WAAW,KAAK,sBAAsB,EACzC,OAAM,kBAAkB;GAI1B,MAAM,OAAO,mBADD,MAAM,kBAAkB,KAAK,CACL;AACpC,SAAM,QAAQ,gBAAgB,IAAI,KAAK;AACvC;;AAGF,cAAY,MAAM;EAGlB,MAAM,YAAY,MAAM,MAAM,SAAS;AACvC,UAAQ,OAAO,MACb,yBAAyB,SAAS,GAAG,MAAM,WAAW,MAAM,UAAU,WAAW,CAAC,IAAI,UAAU,KAAK,KACtG;;AAIH,KAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;AAGH,QAAO;;AAOT,eAAsB,oBACpB,SACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;AAEpF,MAAK,MAAM,WAAW,UAAU;AAC9B,QAAM,kBAAkB;EACxB,MAAM,IAAI,MAAM,wBAAwB,SAAS,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzG,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ"}

package/dist/kg-B5ysyRLC.mjs

@@ -0,0 +1,94 @@
+//#region src/memory/kg.ts
+function rowToTriple(row) {
+	return {
+		id: row.id,
+		subject: row.subject,
+		predicate: row.predicate,
+		object: row.object,
+		project_id: row.project_id,
+		source_session: row.source_session,
+		valid_from: new Date(row.valid_from),
+		valid_to: row.valid_to ? new Date(row.valid_to) : void 0,
+		confidence: row.confidence,
+		created_at: new Date(row.created_at)
+	};
+}
+/**
+* Add a new triple to the knowledge graph.
+* Returns the inserted triple.
+*/
+async function kgAdd(pool, params) {
+	const confidence = params.confidence ?? "EXTRACTED";
+	return rowToTriple((await pool.query(`INSERT INTO kg_triples
+       (subject, predicate, object, project_id, source_session, confidence)
+     VALUES ($1, $2, $3, $4, $5, $6)
+     RETURNING *`, [
+		params.subject,
+		params.predicate,
+		params.object,
+		params.project_id ?? null,
+		params.source_session ?? null,
+		confidence
+	])).rows[0]);
+}
+/**
+* Query triples by subject, predicate, object, and/or project.
+* Supports point-in-time queries via as_of.
+* By default only returns currently-valid triples (valid_to IS NULL).
+*/
+async function kgQuery(pool, params) {
+	const conditions = [];
+	const values = [];
+	let idx = 1;
+	if (params.subject !== void 0) {
+		conditions.push(`subject = $${idx++}`);
+		values.push(params.subject);
+	}
+	if (params.predicate !== void 0) {
+		conditions.push(`predicate = $${idx++}`);
+		values.push(params.predicate);
+	}
+	if (params.object !== void 0) {
+		conditions.push(`object = $${idx++}`);
+		values.push(params.object);
+	}
+	if (params.project_id !== void 0) {
+		conditions.push(`project_id = $${idx++}`);
+		values.push(params.project_id);
+	}
+	if (params.as_of !== void 0) {
+		conditions.push(`valid_from <= $${idx++}`);
+		values.push(params.as_of);
+		conditions.push(`(valid_to IS NULL OR valid_to > $${idx++})`);
+		values.push(params.as_of);
+	} else if (!params.include_invalidated) conditions.push(`valid_to IS NULL`);
+	const where = conditions.length > 0 ? `WHERE ${conditions.join(" AND ")}` : "";
+	return (await pool.query(`SELECT * FROM kg_triples ${where} ORDER BY valid_from DESC`, values)).rows.map(rowToTriple);
+}
+/**
+* Invalidate a triple by setting valid_to = NOW().
+* Does not delete the row — preserves history.
+*/
+async function kgInvalidate(pool, tripleId) {
+	await pool.query(`UPDATE kg_triples SET valid_to = NOW() WHERE id = $1 AND valid_to IS NULL`, [tripleId]);
+}
+/**
+* Find contradictions: cases where the same (subject, predicate) pair has
+* multiple currently-valid objects.
+*/
+async function kgContradictions(pool, subject) {
+	return (await pool.query(`SELECT subject, predicate, array_agg(object ORDER BY object) AS objects
+     FROM kg_triples
+     WHERE subject = $1
+       AND valid_to IS NULL
+     GROUP BY subject, predicate
+     HAVING COUNT(*) > 1`, [subject])).rows.map((row) => ({
+		subject: row.subject,
+		predicate: row.predicate,
+		objects: row.objects
+	}));
+}
+
+//#endregion
+export { kgQuery as i, kgContradictions as n, kgInvalidate as r, kgAdd as t };
+//# sourceMappingURL=kg-B5ysyRLC.mjs.map

package/dist/kg-B5ysyRLC.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"kg-B5ysyRLC.mjs","names":[],"sources":["../src/memory/kg.ts"],"sourcesContent":["/**\n * Temporal Knowledge Graph — kg_triples CRUD layer.\n *\n * Uses the Postgres connection pool from the storage backend.\n * Triples are time-scoped: valid_from/valid_to enable point-in-time queries.\n * Invalidation sets valid_to = NOW() instead of deleting rows.\n */\n\nimport type { Pool } from \"pg\";\n\n// ---------------------------------------------------------------------------\n// Types\n// ---------------------------------------------------------------------------\n\nexport interface KgTriple {\n  id: number;\n  subject: string;\n  predicate: string;\n  object: string;\n  project_id?: number;\n  source_session?: string;\n  valid_from: Date;\n  valid_to?: Date;\n  confidence: \"EXTRACTED\" | \"INFERRED\" | \"AMBIGUOUS\";\n  created_at: Date;\n}\n\nexport interface KgAddParams {\n  subject: string;\n  predicate: string;\n  object: string;\n  project_id?: number;\n  source_session?: string;\n  confidence?: \"EXTRACTED\" | \"INFERRED\" | \"AMBIGUOUS\";\n}\n\nexport interface KgQueryParams {\n  subject?: string;\n  predicate?: string;\n  object?: string;\n  project_id?: number;\n  as_of?: Date;\n  include_invalidated?: boolean;\n}\n\nexport interface KgContradiction {\n  subject: string;\n  predicate: string;\n  objects: string[];\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\nfunction rowToTriple(row: Record<string, unknown>): KgTriple {\n  return {\n    id: row.id as number,\n    subject: row.subject as string,\n    predicate: row.predicate as string,\n    object: row.object as string,\n    project_id: row.project_id as number | undefined,\n    source_session: row.source_session as string | undefined,\n    valid_from: new Date(row.valid_from as string),\n    valid_to: row.valid_to ? new Date(row.valid_to as string) : undefined,\n    confidence: row.confidence as \"EXTRACTED\" | \"INFERRED\" | \"AMBIGUOUS\",\n    created_at: new Date(row.created_at as string),\n  };\n}\n\n// ---------------------------------------------------------------------------\n// Core operations\n// ---------------------------------------------------------------------------\n\n/**\n * Add a new triple to the knowledge graph.\n * Returns the inserted triple.\n */\nexport async function kgAdd(pool: Pool, params: KgAddParams): Promise<KgTriple> {\n  const confidence = params.confidence ?? \"EXTRACTED\";\n  const result = await pool.query<Record<string, unknown>>(\n    `INSERT INTO kg_triples\n       (subject, predicate, object, project_id, source_session, confidence)\n     VALUES ($1, $2, $3, $4, $5, $6)\n     RETURNING *`,\n    [\n      params.subject,\n      params.predicate,\n      params.object,\n      params.project_id ?? null,\n      params.source_session ?? null,\n      confidence,\n    ]\n  );\n  return rowToTriple(result.rows[0]);\n}\n\n/**\n * Query triples by subject, predicate, object, and/or project.\n * Supports point-in-time queries via as_of.\n * By default only returns currently-valid triples (valid_to IS NULL).\n */\nexport async function kgQuery(pool: Pool, params: KgQueryParams): Promise<KgTriple[]> {\n  const conditions: string[] = [];\n  const values: unknown[] = [];\n  let idx = 1;\n\n  if (params.subject !== undefined) {\n    conditions.push(`subject = $${idx++}`);\n    values.push(params.subject);\n  }\n  if (params.predicate !== undefined) {\n    conditions.push(`predicate = $${idx++}`);\n    values.push(params.predicate);\n  }\n  if (params.object !== undefined) {\n    conditions.push(`object = $${idx++}`);\n    values.push(params.object);\n  }\n  if (params.project_id !== undefined) {\n    conditions.push(`project_id = $${idx++}`);\n    values.push(params.project_id);\n  }\n\n  if (params.as_of !== undefined) {\n    // Valid at the given timestamp: started before or at as_of, and not yet ended\n    conditions.push(`valid_from <= $${idx++}`);\n    values.push(params.as_of);\n    conditions.push(`(valid_to IS NULL OR valid_to > $${idx++})`);\n    values.push(params.as_of);\n  } else if (!params.include_invalidated) {\n    // Default: only currently-valid (no valid_to set)\n    conditions.push(`valid_to IS NULL`);\n  }\n\n  const where = conditions.length > 0 ? `WHERE ${conditions.join(\" AND \")}` : \"\";\n  const result = await pool.query<Record<string, unknown>>(\n    `SELECT * FROM kg_triples ${where} ORDER BY valid_from DESC`,\n    values\n  );\n  return result.rows.map(rowToTriple);\n}\n\n/**\n * Invalidate a triple by setting valid_to = NOW().\n * Does not delete the row — preserves history.\n */\nexport async function kgInvalidate(pool: Pool, tripleId: number): Promise<void> {\n  await pool.query(\n    `UPDATE kg_triples SET valid_to = NOW() WHERE id = $1 AND valid_to IS NULL`,\n    [tripleId]\n  );\n}\n\n/**\n * Find contradictions: cases where the same (subject, predicate) pair has\n * multiple currently-valid objects.\n */\nexport async function kgContradictions(\n  pool: Pool,\n  subject: string\n): Promise<KgContradiction[]> {\n  const result = await pool.query<{ subject: string; predicate: string; objects: string[] }>(\n    `SELECT subject, predicate, array_agg(object ORDER BY object) AS objects\n     FROM kg_triples\n     WHERE subject = $1\n       AND valid_to IS NULL\n     GROUP BY subject, predicate\n     HAVING COUNT(*) > 1`,\n    [subject]\n  );\n  return result.rows.map((row) => ({\n    subject: row.subject,\n    predicate: row.predicate,\n    objects: row.objects,\n  }));\n}\n"],"mappings":";AAuDA,SAAS,YAAY,KAAwC;AAC3D,QAAO;EACL,IAAI,IAAI;EACR,SAAS,IAAI;EACb,WAAW,IAAI;EACf,QAAQ,IAAI;EACZ,YAAY,IAAI;EAChB,gBAAgB,IAAI;EACpB,YAAY,IAAI,KAAK,IAAI,WAAqB;EAC9C,UAAU,IAAI,WAAW,IAAI,KAAK,IAAI,SAAmB,GAAG;EAC5D,YAAY,IAAI;EAChB,YAAY,IAAI,KAAK,IAAI,WAAqB;EAC/C;;;;;;AAWH,eAAsB,MAAM,MAAY,QAAwC;CAC9E,MAAM,aAAa,OAAO,cAAc;AAexC,QAAO,aAdQ,MAAM,KAAK,MACxB;;;mBAIA;EACE,OAAO;EACP,OAAO;EACP,OAAO;EACP,OAAO,cAAc;EACrB,OAAO,kBAAkB;EACzB;EACD,CACF,EACyB,KAAK,GAAG;;;;;;;AAQpC,eAAsB,QAAQ,MAAY,QAA4C;CACpF,MAAM,aAAuB,EAAE;CAC/B,MAAM,SAAoB,EAAE;CAC5B,IAAI,MAAM;AAEV,KAAI,OAAO,YAAY,QAAW;AAChC,aAAW,KAAK,cAAc,QAAQ;AACtC,SAAO,KAAK,OAAO,QAAQ;;AAE7B,KAAI,OAAO,cAAc,QAAW;AAClC,aAAW,KAAK,gBAAgB,QAAQ;AACxC,SAAO,KAAK,OAAO,UAAU;;AAE/B,KAAI,OAAO,WAAW,QAAW;AAC/B,aAAW,KAAK,aAAa,QAAQ;AACrC,SAAO,KAAK,OAAO,OAAO;;AAE5B,KAAI,OAAO,eAAe,QAAW;AACnC,aAAW,KAAK,iBAAiB,QAAQ;AACzC,SAAO,KAAK,OAAO,WAAW;;AAGhC,KAAI,OAAO,UAAU,QAAW;AAE9B,aAAW,KAAK,kBAAkB,QAAQ;AAC1C,SAAO,KAAK,OAAO,MAAM;AACzB,aAAW,KAAK,oCAAoC,MAAM,GAAG;AAC7D,SAAO,KAAK,OAAO,MAAM;YAChB,CAAC,OAAO,oBAEjB,YAAW,KAAK,mBAAmB;CAGrC,MAAM,QAAQ,WAAW,SAAS,IAAI,SAAS,WAAW,KAAK,QAAQ,KAAK;AAK5E,SAJe,MAAM,KAAK,MACxB,4BAA4B,MAAM,4BAClC,OACD,EACa,KAAK,IAAI,YAAY;;;;;;AAOrC,eAAsB,aAAa,MAAY,UAAiC;AAC9E,OAAM,KAAK,MACT,6EACA,CAAC,SAAS,CACX;;;;;;AAOH,eAAsB,iBACpB,MACA,SAC4B;AAU5B,SATe,MAAM,KAAK,MACxB;;;;;2BAMA,CAAC,QAAQ,CACV,EACa,KAAK,KAAK,SAAS;EAC/B,SAAS,IAAI;EACb,WAAW,IAAI;EACf,SAAS,IAAI;EACd,EAAE"}

package/dist/kg-extraction-BlGM40q7.mjs

@@ -0,0 +1,211 @@
+import { i as kgQuery, r as kgInvalidate, t as kgAdd } from "./kg-B5ysyRLC.mjs";
+import { existsSync } from "node:fs";
+import { homedir } from "node:os";
+import { join } from "node:path";
+
+//#region src/daemon/templates/triple-extraction-prompt.ts
+/**
+* triple-extraction-prompt.ts — Prompt template for KG triple extraction.
+*
+* Used by the session-summary-worker to extract structured facts from
+* a completed session summary and store them in the temporal knowledge graph.
+*/
+function buildTripleExtractionPrompt(params) {
+	return `Extract atomic facts from this coding session as JSON triples.
+
+A triple has three parts:
+- subject: the entity being described (project name, person, file, concept)
+- predicate: the relationship (uses, depends_on, version, status, lives_at, decided_to, etc.)
+- object: the value or other entity
+
+Output ONLY a JSON array. Each fact must be verifiable from the session content.
+
+Rules:
+- Be SPECIFIC: "Glidr uses FSRS algorithm" not "the project uses an algorithm"
+- Use snake_case predicates
+- Skip opinions, speculation, and "we should" statements
+- Skip facts already obvious from project metadata (e.g., "PAI is written in TypeScript" if PAI is the project)
+- Maximum 15 triples per session — pick the most important
+- Each triple should be a fact that might be queried later
+
+Example output:
+[
+  {"subject": "Glidr", "predicate": "uses_algorithm", "object": "FSRS"},
+  {"subject": "Glidr", "predicate": "shipped_version", "object": "1.0.5"},
+  {"subject": "Quassl", "predicate": "platform", "object": "iOS"},
+  {"subject": "Matthias", "predicate": "decided_to", "object": "rewrite Quassl in Flutter"}
+]
+
+PROJECT: ${params.projectSlug}
+
+SESSION CONTENT:
+${params.sessionContent}
+
+GIT COMMITS:
+${params.gitLog}
+
+JSON triples:`;
+}
+
+//#endregion
+//#region src/memory/kg-extraction.ts
+/**
+* kg-extraction.ts — Shared KG triple extraction logic.
+*
+* Extracted from session-summary-worker.ts so both the worker and the
+* CLI backfill (`pai kg backfill`) can use the same code path.
+*
+* Provides:
+*   - findClaudeBinary()       — locate the claude CLI
+*   - spawnClaude()            — generic prompt -> response runner (strips ANTHROPIC_API_KEY)
+*   - extractAndStoreTriples() — run the extractor prompt and persist triples to Postgres
+*/
+/**
+* Find the `claude` CLI binary. Checks common installation locations first
+* (launchd PATH is minimal so bare "claude" often won't resolve).
+*/
+function findClaudeBinary() {
+	const candidates = [
+		join(homedir(), ".local", "bin", "claude"),
+		join(homedir(), ".claude", "local", "claude"),
+		"/usr/local/bin/claude",
+		"/opt/homebrew/bin/claude"
+	];
+	for (const candidate of candidates) try {
+		if (existsSync(candidate)) return candidate;
+	} catch {}
+	return "claude";
+}
+const CLAUDE_TIMEOUT_MS = {
+	haiku: 6e4,
+	sonnet: 12e4,
+	opus: 3e5
+};
+/**
+* Spawn the claude CLI with a prompt on stdin and return stdout.
+*
+* IMPORTANT: ANTHROPIC_API_KEY is stripped from the spawned environment so
+* the CLI uses the user's Max plan (free) instead of billing the API key.
+*/
+async function spawnClaude(prompt, model = "sonnet") {
+	const claudeBin = findClaudeBinary();
+	if (!claudeBin) {
+		process.stderr.write("[kg-extraction] claude CLI not found.\n");
+		return null;
+	}
+	const { spawn } = await import("node:child_process");
+	return new Promise((resolve) => {
+		let timer = null;
+		const { ANTHROPIC_API_KEY: _drop, ...envWithoutApiKey } = process.env;
+		const child = spawn(claudeBin, [
+			"--model",
+			model,
+			"-p",
+			"--no-session-persistence"
+		], {
+			env: envWithoutApiKey,
+			stdio: [
+				"pipe",
+				"pipe",
+				"pipe"
+			]
+		});
+		let stdout = "";
+		let stderr = "";
+		child.stdout.on("data", (chunk) => {
+			stdout += chunk.toString();
+		});
+		child.stderr.on("data", (chunk) => {
+			stderr += chunk.toString();
+		});
+		child.on("error", (err) => {
+			if (timer) {
+				clearTimeout(timer);
+				timer = null;
+			}
+			process.stderr.write(`[kg-extraction] ${model} spawn error: ${err.message}\n`);
+			resolve(null);
+		});
+		child.on("close", (code) => {
+			if (timer) {
+				clearTimeout(timer);
+				timer = null;
+			}
+			if (code !== 0) {
+				process.stderr.write(`[kg-extraction] ${model} exited ${code}: ${stderr.slice(0, 300)}\n`);
+				resolve(null);
+			} else resolve(stdout.trim() || null);
+		});
+		timer = setTimeout(() => {
+			process.stderr.write(`[kg-extraction] ${model} timed out — killing process.\n`);
+			child.kill("SIGTERM");
+			resolve(null);
+		}, CLAUDE_TIMEOUT_MS[model] ?? 12e4);
+		child.stdin.write(prompt);
+		child.stdin.end();
+	});
+}
+/**
+* Extract structured KG triples from a session summary and store them in
+* Postgres. Idempotent: if a (subject, predicate) pair already has the same
+* object, no new row is added; if the object differs, the old triple is
+* invalidated (valid_to = NOW()) and a new one is inserted.
+*
+* Best-effort: per-triple errors are caught and logged but never thrown.
+* Returns a small stats object so callers can report progress.
+*/
+async function extractAndStoreTriples(pool, params) {
+	const stats = {
+		extracted: 0,
+		added: 0,
+		superseded: 0
+	};
+	const jsonOutput = await spawnClaude(buildTripleExtractionPrompt({
+		sessionContent: params.summaryText,
+		projectSlug: params.projectSlug,
+		gitLog: params.gitLog ?? ""
+	}), params.model ?? "sonnet");
+	if (!jsonOutput) return stats;
+	const cleaned = jsonOutput.replace(/^```json\s*/m, "").replace(/^```\s*/m, "").replace(/\s*```$/m, "").trim();
+	let triples;
+	try {
+		triples = JSON.parse(cleaned);
+	} catch (e) {
+		process.stderr.write(`[kg-extraction] JSON parse failed: ${e}\n`);
+		return stats;
+	}
+	if (!Array.isArray(triples)) return stats;
+	stats.extracted = triples.length;
+	for (const t of triples) {
+		if (!t.subject || !t.predicate || !t.object) continue;
+		try {
+			const existing = await kgQuery(pool, {
+				subject: t.subject,
+				predicate: t.predicate,
+				project_id: params.projectId ?? void 0
+			});
+			if (existing.find((e) => e.object === t.object && !e.valid_to)) continue;
+			const supersedes = existing.find((e) => e.object !== t.object && !e.valid_to);
+			if (supersedes) {
+				await kgInvalidate(pool, supersedes.id);
+				stats.superseded++;
+			}
+			await kgAdd(pool, {
+				subject: t.subject,
+				predicate: t.predicate,
+				object: t.object,
+				project_id: params.projectId ?? void 0,
+				source_session: params.sessionId,
+				confidence: "EXTRACTED"
+			});
+			stats.added++;
+		} catch (tripleErr) {
+			process.stderr.write(`[kg-extraction] store error (${t.subject}): ${tripleErr}\n`);
+		}
+	}
+	return stats;
+}
+
+//#endregion
+export { extractAndStoreTriples as t };
+//# sourceMappingURL=kg-extraction-BlGM40q7.mjs.map

package/dist/kg-extraction-BlGM40q7.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"kg-extraction-BlGM40q7.mjs","names":[],"sources":["../src/daemon/templates/triple-extraction-prompt.ts","../src/memory/kg-extraction.ts"],"sourcesContent":["/**\n * triple-extraction-prompt.ts — Prompt template for KG triple extraction.\n *\n * Used by the session-summary-worker to extract structured facts from\n * a completed session summary and store them in the temporal knowledge graph.\n */\n\nexport function buildTripleExtractionPrompt(params: {\n  sessionContent: string;\n  projectSlug: string;\n  gitLog: string;\n}): string {\n  return `Extract atomic facts from this coding session as JSON triples.\n\nA triple has three parts:\n- subject: the entity being described (project name, person, file, concept)\n- predicate: the relationship (uses, depends_on, version, status, lives_at, decided_to, etc.)\n- object: the value or other entity\n\nOutput ONLY a JSON array. Each fact must be verifiable from the session content.\n\nRules:\n- Be SPECIFIC: \"Glidr uses FSRS algorithm\" not \"the project uses an algorithm\"\n- Use snake_case predicates\n- Skip opinions, speculation, and \"we should\" statements\n- Skip facts already obvious from project metadata (e.g., \"PAI is written in TypeScript\" if PAI is the project)\n- Maximum 15 triples per session — pick the most important\n- Each triple should be a fact that might be queried later\n\nExample output:\n[\n  {\"subject\": \"Glidr\", \"predicate\": \"uses_algorithm\", \"object\": \"FSRS\"},\n  {\"subject\": \"Glidr\", \"predicate\": \"shipped_version\", \"object\": \"1.0.5\"},\n  {\"subject\": \"Quassl\", \"predicate\": \"platform\", \"object\": \"iOS\"},\n  {\"subject\": \"Matthias\", \"predicate\": \"decided_to\", \"object\": \"rewrite Quassl in Flutter\"}\n]\n\nPROJECT: ${params.projectSlug}\n\nSESSION CONTENT:\n${params.sessionContent}\n\nGIT COMMITS:\n${params.gitLog}\n\nJSON triples:`;\n}\n","/**\n * kg-extraction.ts — Shared KG triple extraction logic.\n *\n * Extracted from session-summary-worker.ts so both the worker and the\n * CLI backfill (`pai kg backfill`) can use the same code path.\n *\n * Provides:\n *   - findClaudeBinary()       — locate the claude CLI\n *   - spawnClaude()            — generic prompt -> response runner (strips ANTHROPIC_API_KEY)\n *   - extractAndStoreTriples() — run the extractor prompt and persist triples to Postgres\n */\n\nimport { existsSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { homedir } from \"node:os\";\nimport type { Pool } from \"pg\";\n\nimport { buildTripleExtractionPrompt } from \"../daemon/templates/triple-extraction-prompt.js\";\nimport { kgAdd, kgQuery, kgInvalidate } from \"./kg.js\";\n\n// ---------------------------------------------------------------------------\n// Claude CLI binary discovery\n// ---------------------------------------------------------------------------\n\n/**\n * Find the `claude` CLI binary. Checks common installation locations first\n * (launchd PATH is minimal so bare \"claude\" often won't resolve).\n */\nexport function findClaudeBinary(): string | null {\n  const candidates = [\n    join(homedir(), \".local\", \"bin\", \"claude\"),\n    join(homedir(), \".claude\", \"local\", \"claude\"),\n    \"/usr/local/bin/claude\",\n    \"/opt/homebrew/bin/claude\",\n  ];\n\n  for (const candidate of candidates) {\n    try {\n      if (existsSync(candidate)) return candidate;\n    } catch { /* skip */ }\n  }\n  return \"claude\";\n}\n\nconst CLAUDE_TIMEOUT_MS: Record<string, number> = {\n  haiku: 60_000,\n  sonnet: 120_000,\n  opus: 300_000,\n};\n\n/**\n * Spawn the claude CLI with a prompt on stdin and return stdout.\n *\n * IMPORTANT: ANTHROPIC_API_KEY is stripped from the spawned environment so\n * the CLI uses the user's Max plan (free) instead of billing the API key.\n */\nexport async function spawnClaude(\n  prompt: string,\n  model: \"haiku\" | \"sonnet\" | \"opus\" = \"sonnet\"\n): Promise<string | null> {\n  const claudeBin = findClaudeBinary();\n  if (!claudeBin) {\n    process.stderr.write(\"[kg-extraction] claude CLI not found.\\n\");\n    return null;\n  }\n\n  const { spawn } = await import(\"node:child_process\");\n\n  return new Promise((resolve) => {\n    let timer: ReturnType<typeof setTimeout> | null = null;\n\n    const { ANTHROPIC_API_KEY: _drop, ...envWithoutApiKey } = process.env;\n    const child = spawn(\n      claudeBin,\n      [\"--model\", model, \"-p\", \"--no-session-persistence\"],\n      { env: envWithoutApiKey, stdio: [\"pipe\", \"pipe\", \"pipe\"] }\n    );\n\n    let stdout = \"\";\n    let stderr = \"\";\n\n    child.stdout.on(\"data\", (chunk: Buffer) => { stdout += chunk.toString(); });\n    child.stderr.on(\"data\", (chunk: Buffer) => { stderr += chunk.toString(); });\n\n    child.on(\"error\", (err: Error) => {\n      if (timer) { clearTimeout(timer); timer = null; }\n      process.stderr.write(`[kg-extraction] ${model} spawn error: ${err.message}\\n`);\n      resolve(null);\n    });\n\n    child.on(\"close\", (code: number | null) => {\n      if (timer) { clearTimeout(timer); timer = null; }\n      if (code !== 0) {\n        process.stderr.write(\n          `[kg-extraction] ${model} exited ${code}: ${stderr.slice(0, 300)}\\n`\n        );\n        resolve(null);\n      } else {\n        resolve(stdout.trim() || null);\n      }\n    });\n\n    timer = setTimeout(() => {\n      process.stderr.write(`[kg-extraction] ${model} timed out — killing process.\\n`);\n      child.kill(\"SIGTERM\");\n      resolve(null);\n    }, CLAUDE_TIMEOUT_MS[model] ?? 120_000);\n\n    child.stdin.write(prompt);\n    child.stdin.end();\n  });\n}\n\n// ---------------------------------------------------------------------------\n// Triple extraction\n// ---------------------------------------------------------------------------\n\nexport interface ExtractTriplesParams {\n  summaryText: string;\n  projectSlug: string;\n  projectId: number | null;\n  sessionId: string;\n  gitLog?: string;\n  model?: \"haiku\" | \"sonnet\" | \"opus\";\n}\n\nexport interface ExtractTriplesResult {\n  extracted: number;\n  added: number;\n  superseded: number;\n}\n\n/**\n * Extract structured KG triples from a session summary and store them in\n * Postgres. Idempotent: if a (subject, predicate) pair already has the same\n * object, no new row is added; if the object differs, the old triple is\n * invalidated (valid_to = NOW()) and a new one is inserted.\n *\n * Best-effort: per-triple errors are caught and logged but never thrown.\n * Returns a small stats object so callers can report progress.\n */\nexport async function extractAndStoreTriples(\n  pool: Pool,\n  params: ExtractTriplesParams\n): Promise<ExtractTriplesResult> {\n  const stats: ExtractTriplesResult = { extracted: 0, added: 0, superseded: 0 };\n\n  const prompt = buildTripleExtractionPrompt({\n    sessionContent: params.summaryText,\n    projectSlug: params.projectSlug,\n    gitLog: params.gitLog ?? \"\",\n  });\n\n  const jsonOutput = await spawnClaude(prompt, params.model ?? \"sonnet\");\n  if (!jsonOutput) return stats;\n\n  // Strip markdown code fences if Claude wrapped the JSON\n  const cleaned = jsonOutput\n    .replace(/^```json\\s*/m, \"\")\n    .replace(/^```\\s*/m, \"\")\n    .replace(/\\s*```$/m, \"\")\n    .trim();\n\n  let triples: Array<{ subject: string; predicate: string; object: string }>;\n  try {\n    triples = JSON.parse(cleaned);\n  } catch (e) {\n    process.stderr.write(`[kg-extraction] JSON parse failed: ${e}\\n`);\n    return stats;\n  }\n\n  if (!Array.isArray(triples)) return stats;\n  stats.extracted = triples.length;\n\n  for (const t of triples) {\n    if (!t.subject || !t.predicate || !t.object) continue;\n\n    try {\n      const existing = await kgQuery(pool, {\n        subject: t.subject,\n        predicate: t.predicate,\n        project_id: params.projectId ?? undefined,\n      });\n\n      // If an identical (subject, predicate, object) is already valid, skip — idempotent\n      const alreadyValid = existing.find((e) => e.object === t.object && !e.valid_to);\n      if (alreadyValid) continue;\n\n      // Invalidate any superseded triple (same subject+predicate, different object)\n      const supersedes = existing.find((e) => e.object !== t.object && !e.valid_to);\n      if (supersedes) {\n        await kgInvalidate(pool, supersedes.id);\n        stats.superseded++;\n      }\n\n      await kgAdd(pool, {\n        subject: t.subject,\n        predicate: t.predicate,\n        object: t.object,\n        project_id: params.projectId ?? undefined,\n        source_session: params.sessionId,\n        confidence: \"EXTRACTED\",\n      });\n      stats.added++;\n    } catch (tripleErr) {\n      process.stderr.write(`[kg-extraction] store error (${t.subject}): ${tripleErr}\\n`);\n    }\n  }\n\n  return stats;\n}\n"],"mappings":";;;;;;;;;;;;AAOA,SAAgB,4BAA4B,QAIjC;AACT,QAAO;;;;;;;;;;;;;;;;;;;;;;;;;WAyBE,OAAO,YAAY;;;EAG5B,OAAO,eAAe;;;EAGtB,OAAO,OAAO;;;;;;;;;;;;;;;;;;;;;;ACfhB,SAAgB,mBAAkC;CAChD,MAAM,aAAa;EACjB,KAAK,SAAS,EAAE,UAAU,OAAO,SAAS;EAC1C,KAAK,SAAS,EAAE,WAAW,SAAS,SAAS;EAC7C;EACA;EACD;AAED,MAAK,MAAM,aAAa,WACtB,KAAI;AACF,MAAI,WAAW,UAAU,CAAE,QAAO;SAC5B;AAEV,QAAO;;AAGT,MAAM,oBAA4C;CAChD,OAAO;CACP,QAAQ;CACR,MAAM;CACP;;;;;;;AAQD,eAAsB,YACpB,QACA,QAAqC,UACb;CACxB,MAAM,YAAY,kBAAkB;AACpC,KAAI,CAAC,WAAW;AACd,UAAQ,OAAO,MAAM,0CAA0C;AAC/D,SAAO;;CAGT,MAAM,EAAE,UAAU,MAAM,OAAO;AAE/B,QAAO,IAAI,SAAS,YAAY;EAC9B,IAAI,QAA8C;EAElD,MAAM,EAAE,mBAAmB,OAAO,GAAG,qBAAqB,QAAQ;EAClE,MAAM,QAAQ,MACZ,WACA;GAAC;GAAW;GAAO;GAAM;GAA2B,EACpD;GAAE,KAAK;GAAkB,OAAO;IAAC;IAAQ;IAAQ;IAAO;GAAE,CAC3D;EAED,IAAI,SAAS;EACb,IAAI,SAAS;AAEb,QAAM,OAAO,GAAG,SAAS,UAAkB;AAAE,aAAU,MAAM,UAAU;IAAI;AAC3E,QAAM,OAAO,GAAG,SAAS,UAAkB;AAAE,aAAU,MAAM,UAAU;IAAI;AAE3E,QAAM,GAAG,UAAU,QAAe;AAChC,OAAI,OAAO;AAAE,iBAAa,MAAM;AAAE,YAAQ;;AAC1C,WAAQ,OAAO,MAAM,mBAAmB,MAAM,gBAAgB,IAAI,QAAQ,IAAI;AAC9E,WAAQ,KAAK;IACb;AAEF,QAAM,GAAG,UAAU,SAAwB;AACzC,OAAI,OAAO;AAAE,iBAAa,MAAM;AAAE,YAAQ;;AAC1C,OAAI,SAAS,GAAG;AACd,YAAQ,OAAO,MACb,mBAAmB,MAAM,UAAU,KAAK,IAAI,OAAO,MAAM,GAAG,IAAI,CAAC,IAClE;AACD,YAAQ,KAAK;SAEb,SAAQ,OAAO,MAAM,IAAI,KAAK;IAEhC;AAEF,UAAQ,iBAAiB;AACvB,WAAQ,OAAO,MAAM,mBAAmB,MAAM,iCAAiC;AAC/E,SAAM,KAAK,UAAU;AACrB,WAAQ,KAAK;KACZ,kBAAkB,UAAU,KAAQ;AAEvC,QAAM,MAAM,MAAM,OAAO;AACzB,QAAM,MAAM,KAAK;GACjB;;;;;;;;;;;AA+BJ,eAAsB,uBACpB,MACA,QAC+B;CAC/B,MAAM,QAA8B;EAAE,WAAW;EAAG,OAAO;EAAG,YAAY;EAAG;CAQ7E,MAAM,aAAa,MAAM,YANV,4BAA4B;EACzC,gBAAgB,OAAO;EACvB,aAAa,OAAO;EACpB,QAAQ,OAAO,UAAU;EAC1B,CAAC,EAE2C,OAAO,SAAS,SAAS;AACtE,KAAI,CAAC,WAAY,QAAO;CAGxB,MAAM,UAAU,WACb,QAAQ,gBAAgB,GAAG,CAC3B,QAAQ,YAAY,GAAG,CACvB,QAAQ,YAAY,GAAG,CACvB,MAAM;CAET,IAAI;AACJ,KAAI;AACF,YAAU,KAAK,MAAM,QAAQ;UACtB,GAAG;AACV,UAAQ,OAAO,MAAM,sCAAsC,EAAE,IAAI;AACjE,SAAO;;AAGT,KAAI,CAAC,MAAM,QAAQ,QAAQ,CAAE,QAAO;AACpC,OAAM,YAAY,QAAQ;AAE1B,MAAK,MAAM,KAAK,SAAS;AACvB,MAAI,CAAC,EAAE,WAAW,CAAC,EAAE,aAAa,CAAC,EAAE,OAAQ;AAE7C,MAAI;GACF,MAAM,WAAW,MAAM,QAAQ,MAAM;IACnC,SAAS,EAAE;IACX,WAAW,EAAE;IACb,YAAY,OAAO,aAAa;IACjC,CAAC;AAIF,OADqB,SAAS,MAAM,MAAM,EAAE,WAAW,EAAE,UAAU,CAAC,EAAE,SAAS,CAC7D;GAGlB,MAAM,aAAa,SAAS,MAAM,MAAM,EAAE,WAAW,EAAE,UAAU,CAAC,EAAE,SAAS;AAC7E,OAAI,YAAY;AACd,UAAM,aAAa,MAAM,WAAW,GAAG;AACvC,UAAM;;AAGR,SAAM,MAAM,MAAM;IAChB,SAAS,EAAE;IACX,WAAW,EAAE;IACb,QAAQ,EAAE;IACV,YAAY,OAAO,aAAa;IAChC,gBAAgB,OAAO;IACvB,YAAY;IACb,CAAC;AACF,SAAM;WACC,WAAW;AAClB,WAAQ,OAAO,MAAM,gCAAgC,EAAE,QAAQ,KAAK,UAAU,IAAI;;;AAItF,QAAO"}

package/dist/{latent-ideas-bTJo6Omd.mjs → latent-ideas-DvWBRHsy.mjs}

@@ -1,6 +1,6 @@
 import "./embeddings-DGRAPAYb.mjs";
 import { n as TITLE_STOP_WORDS } from "./stop-words-BaMEGVeY.mjs";
-import { t as zettelThemes } from "./themes-BvYF0W8T.mjs";
+import { t as zettelThemes } from "./themes-9jxFn3Rf.mjs";
 import { mkdirSync, writeFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 
@@ -188,4 +188,4 @@ function handleIdeaMaterialize(params, vaultPath) {
 
 //#endregion
 export { handleGraphLatentIdeas, handleIdeaMaterialize };
-//# sourceMappingURL=latent-ideas-bTJo6Omd.mjs.map
+//# sourceMappingURL=latent-ideas-DvWBRHsy.mjs.map