@tekmidian/pai 0.9.0 → 0.9.2

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

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

@@ -1 +1 @@
-{"version":3,"file":"latent-ideas-bTJo6Omd.mjs","names":[],"sources":["../src/graph/latent-ideas.ts"],"sourcesContent":["/**\n * latent-ideas.ts — graph_latent_ideas and idea_materialize endpoint handlers\n *\n * \"Latent ideas\" are recurring themes in the vault that exist as embedding\n * clusters but have NO dedicated note written about them yet.  PAI surfaces\n * these by running the same agglomerative clustering used by graph_clusters /\n * zettelThemes and then filtering OUT any cluster whose label is well-matched\n * by an existing note title.\n *\n * The materialize endpoint writes a new Markdown note to the vault filesystem\n * and returns its content so the plugin can open it immediately.\n */\n\nimport { mkdirSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport { TITLE_STOP_WORDS } from \"../utils/stop-words.js\";\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport { zettelThemes } from \"../zettelkasten/themes.js\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphLatentIdeasParams {\n  project_id: number;\n  /** Minimum notes in a cluster (default: 3) */\n  min_cluster_size?: number;\n  /** Cap on returned ideas (default: 15) */\n  max_ideas?: number;\n  /** How far back to look in days (default: 180) */\n  lookback_days?: number;\n  /** Cosine similarity clustering threshold (default: 0.65) */\n  similarity_threshold?: number;\n}\n\nexport interface LatentIdeaSourceNote {\n  vault_path: string;\n  title: string;\n  /** How strongly this note relates to the theme (0-1) */\n  relevance: number;\n}\n\nexport interface LatentIdea {\n  id: number;\n  /** Auto-generated cluster label from zettelThemes */\n  label: string;\n  /** Number of notes touching this theme */\n  size: number;\n  /** 0-1, how likely this is a real coherent idea */\n  confidence: number;\n  /** Notes that contribute to this theme */\n  source_notes: LatentIdeaSourceNote[];\n  /** Cleaned-up version of label for a potential note title */\n  suggested_title: string;\n  /** Most common folder among source notes */\n  suggested_folder: string;\n  /** Number of distinct session date-folders (e.g. \"2026/03\") touching this theme */\n  sessions_count: number;\n}\n\nexport interface GraphLatentIdeasResult {\n  ideas: LatentIdea[];\n  total_clusters_analyzed: number;\n  /** How many clusters already have a matching note (excluded from results) */\n  materialized_count: number;\n}\n\n// ---------------------------------------------------------------------------\n// Materialize params / result\n// ---------------------------------------------------------------------------\n\nexport interface IdeaMaterializeParams {\n  idea_label: string;\n  /** User-chosen title for the new note */\n  title: string;\n  /** Vault-relative folder path where the note should be created */\n  folder: string;\n  /** Vault-relative paths of the source notes to link from the new note */\n  source_paths: string[];\n  project_id: number;\n}\n\nexport interface IdeaMaterializeResult {\n  /** Vault-relative path of the created note */\n  vault_path: string;\n  /** Generated markdown content */\n  content: string;\n  /** Number of wikilinks inserted */\n  links_created: number;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: check if a cluster already has a matching note\n// ---------------------------------------------------------------------------\n\n/**\n * Returns true when any existing vault note title closely matches the cluster\n * label — meaning a dedicated note already exists for this topic.\n *\n * Matching strategy (simple, fast, no embeddings needed):\n *   1. Lowercase both sides and split into words.\n *   2. Remove stop words from the label words.\n *   3. If ≥ 60% of the significant label words appear in a note title → match.\n */\n// TITLE_STOP_WORDS imported from utils/stop-words.ts\n\nfunction labelMatchesTitle(label: string, title: string): boolean {\n  const labelWords = label\n    .toLowerCase()\n    .split(/[\\s\\-_/]+/)\n    .filter((w) => w.length > 2 && !TITLE_STOP_WORDS.has(w));\n\n  if (labelWords.length === 0) return false;\n\n  const titleLower = title.toLowerCase();\n  const matchCount = labelWords.filter((w) => titleLower.includes(w)).length;\n  return matchCount / labelWords.length >= 0.6;\n}\n\n/**\n * Check whether any note indexed in the vault has a title matching the label.\n * Fetches all vault file rows via StorageBackend for efficiency.\n */\nasync function clusterHasMatchingNote(\n  backend: StorageBackend,\n  label: string,\n  notePaths: string[]\n): Promise<boolean> {\n  // First check the notes already in the cluster themselves — if any cluster\n  // member's title matches the label it IS the index note → materialized.\n  const pathSet = new Set(notePaths);\n\n  // Fetch all vault files (bounded — vault rarely > 50k notes)\n  const rows = await backend.getAllVaultFiles();\n\n  for (const row of rows) {\n    if (!row.title) continue;\n    // Skip notes already counted inside the cluster — they don't count as\n    // \"dedicated notes\"; we only skip a cluster if a SEPARATE note exists.\n    if (pathSet.has(row.vaultPath)) continue;\n    if (labelMatchesTitle(label, row.title)) return true;\n  }\n  return false;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: generate a clean suggested title\n// ---------------------------------------------------------------------------\n\nfunction toSuggestedTitle(label: string): string {\n  // Remove leading/trailing whitespace, capitalize each word, remove stop words\n  // that are all-lowercase at the start of the title.\n  const words = label\n    .trim()\n    .split(/\\s+/)\n    .map((w, i) => {\n      const lower = w.toLowerCase();\n      // Drop leading stop words (but keep if they're the only word)\n      if (i === 0 && TITLE_STOP_WORDS.has(lower) && label.trim().split(/\\s+/).length > 1) {\n        return \"\";\n      }\n      return w.charAt(0).toUpperCase() + w.slice(1);\n    })\n    .filter(Boolean);\n\n  return words.join(\" \") || label;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: find most common folder\n// ---------------------------------------------------------------------------\n\nfunction mostCommonFolder(vaultPaths: string[]): string {\n  const counts = new Map<string, number>();\n  for (const p of vaultPaths) {\n    const parts = p.split(\"/\");\n    const folder = parts.length > 1 ? parts.slice(0, -1).join(\"/\") : \"\";\n    counts.set(folder, (counts.get(folder) ?? 0) + 1);\n  }\n\n  let best = \"\";\n  let bestCount = 0;\n  for (const [folder, count] of counts) {\n    if (count > bestCount) {\n      bestCount = count;\n      best = folder;\n    }\n  }\n  return best;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: count distinct session date-folders\n// ---------------------------------------------------------------------------\n\n/**\n * Heuristic: vault notes are often stored in date-based folders like\n * \"2026/03/15\" or \"Daily/2026-03\".  We extract the first numeric path\n * segment that looks like a year (2020-2030) and group by year+month.\n *\n * Falls back to counting distinct top-level folders.\n */\nfunction countDistinctSessions(vaultPaths: string[]): number {\n  const sessions = new Set<string>();\n  const yearMonthRe = /\\b(202\\d)\\D?(0[1-9]|1[0-2])\\b/;\n\n  for (const p of vaultPaths) {\n    const m = yearMonthRe.exec(p);\n    if (m) {\n      sessions.add(`${m[1]}-${m[2]}`);\n    } else {\n      // Fallback: use top-level folder as a proxy for \"session bucket\"\n      const topFolder = p.split(\"/\")[0];\n      sessions.add(topFolder);\n    }\n  }\n  return sessions.size;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: calculate confidence score\n// ---------------------------------------------------------------------------\n\n/**\n * Confidence combines:\n *  - Cluster size (normalized, capped at 20 for max contribution)\n *  - Folder diversity (0-1 already)\n *  - Sessions count (normalized, capped at 5)\n *\n * Formula: 0.4 * sizeScore + 0.35 * folderDiversity + 0.25 * sessionScore\n */\nfunction calcConfidence(\n  size: number,\n  folderDiversity: number,\n  sessionsCount: number\n): number {\n  const sizeScore = Math.min(size / 20, 1.0);\n  const sessionScore = Math.min(sessionsCount / 5, 1.0);\n  const raw = 0.4 * sizeScore + 0.35 * folderDiversity + 0.25 * sessionScore;\n  return Math.round(raw * 100) / 100;\n}\n\n// ---------------------------------------------------------------------------\n// Main handler: graph_latent_ideas\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphLatentIdeas(\n  backend: StorageBackend,\n  params: GraphLatentIdeasParams\n): Promise<GraphLatentIdeasResult> {\n  const minClusterSize = params.min_cluster_size ?? 3;\n  const maxIdeas = params.max_ideas ?? 15;\n  const lookbackDays = params.lookback_days ?? 180;\n  const similarityThreshold = params.similarity_threshold ?? 0.65;\n\n  const { project_id: vaultProjectId } = params;\n  if (!vaultProjectId) {\n    throw new Error(\n      \"graph_latent_ideas: project_id is required (pass the vault project's numeric ID)\"\n    );\n  }\n\n  // Run the same clustering algorithm used by graph_clusters\n  const themeResult = await zettelThemes(backend, {\n    vaultProjectId,\n    lookbackDays,\n    minClusterSize,\n    maxThemes: maxIdeas * 3, // Over-fetch — many will be filtered as materialized\n    similarityThreshold,\n  });\n\n  const ideas: LatentIdea[] = [];\n  let materializedCount = 0;\n\n  for (const theme of themeResult.themes) {\n    const notePaths = theme.notes.map((n) => n.path);\n\n    // Check if a dedicated note already exists for this theme\n    if (await clusterHasMatchingNote(backend, theme.label, notePaths)) {\n      materializedCount++;\n      continue;\n    }\n\n    // This is a latent idea — no dedicated note exists yet\n    const suggestedFolder = mostCommonFolder(notePaths);\n    const sessionsCount = countDistinctSessions(notePaths);\n    const confidence = calcConfidence(theme.size, theme.folderDiversity, sessionsCount);\n\n    // Build source notes with relevance scores\n    // Relevance is approximated by position in cluster (centroid-closest first)\n    // zettelThemes returns notes in no guaranteed order; assign uniform relevance\n    // decreasing from 1.0 to 0.5 across the list.\n    const sourceNotes: LatentIdeaSourceNote[] = theme.notes.map((n, idx) => ({\n      vault_path: n.path,\n      title: n.title ?? n.path.split(\"/\").pop()?.replace(/\\.md$/i, \"\") ?? n.path,\n      relevance: Math.round((1.0 - (idx / Math.max(theme.notes.length - 1, 1)) * 0.5) * 100) / 100,\n    }));\n\n    ideas.push({\n      id: theme.id,\n      label: theme.label,\n      size: theme.size,\n      confidence,\n      source_notes: sourceNotes,\n      suggested_title: toSuggestedTitle(theme.label),\n      suggested_folder: suggestedFolder,\n      sessions_count: sessionsCount,\n    });\n\n    if (ideas.length >= maxIdeas) break;\n  }\n\n  // Sort by confidence descending\n  ideas.sort((a, b) => b.confidence - a.confidence);\n\n  return {\n    ideas,\n    total_clusters_analyzed: themeResult.themes.length + materializedCount,\n    materialized_count: materializedCount,\n  };\n}\n\n// ---------------------------------------------------------------------------\n// Materialize handler: idea_materialize\n// ---------------------------------------------------------------------------\n\nexport function handleIdeaMaterialize(\n  params: IdeaMaterializeParams,\n  vaultPath: string\n): IdeaMaterializeResult {\n  const { idea_label, title, folder, source_paths } = params;\n\n  // Sanitize filename: replace characters illegal in filenames\n  const safeTitle = title.replace(/[/\\\\:*?\"<>|]/g, \"-\");\n  const fileName = `${safeTitle}.md`;\n\n  // Vault-relative path (forward slashes, no leading slash)\n  const relFolder = folder.replace(/^\\/+|\\/+$/g, \"\");\n  const vault_path = relFolder ? `${relFolder}/${fileName}` : fileName;\n\n  // Absolute filesystem path\n  const absPath = join(vaultPath, vault_path);\n  const absDir = dirname(absPath);\n\n  // Build wikilinks from source_paths\n  const wikilinks = source_paths\n    .map((p) => {\n      // Derive a display name: filename without extension\n      const name = p.split(\"/\").pop()?.replace(/\\.md$/i, \"\") ?? p;\n      // Relative wikilink — use just the filename (Obsidian resolves by title)\n      return `- [[${name}]]`;\n    })\n    .join(\"\\n\");\n\n  const links_created = source_paths.length;\n\n  const content = [\n    `# ${title}`,\n    \"\",\n    `*Materialized from latent idea: \"${idea_label}\"*`,\n    `*Sources: ${links_created} notes*`,\n    \"\",\n    \"## Related Notes\",\n    \"\",\n    wikilinks || \"*(no source notes)*\",\n    \"\",\n    \"## Notes\",\n    \"\",\n    \"<!-- Add your thoughts about this idea here -->\",\n    \"\",\n  ].join(\"\\n\");\n\n  // Write the file (create parent directories as needed)\n  mkdirSync(absDir, { recursive: true });\n  writeFileSync(absPath, content, \"utf-8\");\n\n  return {\n    vault_path,\n    content,\n    links_created,\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AA0GA,SAAS,kBAAkB,OAAe,OAAwB;CAChE,MAAM,aAAa,MAChB,aAAa,CACb,MAAM,YAAY,CAClB,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,iBAAiB,IAAI,EAAE,CAAC;AAE1D,KAAI,WAAW,WAAW,EAAG,QAAO;CAEpC,MAAM,aAAa,MAAM,aAAa;AAEtC,QADmB,WAAW,QAAQ,MAAM,WAAW,SAAS,EAAE,CAAC,CAAC,SAChD,WAAW,UAAU;;;;;;AAO3C,eAAe,uBACb,SACA,OACA,WACkB;CAGlB,MAAM,UAAU,IAAI,IAAI,UAAU;CAGlC,MAAM,OAAO,MAAM,QAAQ,kBAAkB;AAE7C,MAAK,MAAM,OAAO,MAAM;AACtB,MAAI,CAAC,IAAI,MAAO;AAGhB,MAAI,QAAQ,IAAI,IAAI,UAAU,CAAE;AAChC,MAAI,kBAAkB,OAAO,IAAI,MAAM,CAAE,QAAO;;AAElD,QAAO;;AAOT,SAAS,iBAAiB,OAAuB;AAgB/C,QAbc,MACX,MAAM,CACN,MAAM,MAAM,CACZ,KAAK,GAAG,MAAM;EACb,MAAM,QAAQ,EAAE,aAAa;AAE7B,MAAI,MAAM,KAAK,iBAAiB,IAAI,MAAM,IAAI,MAAM,MAAM,CAAC,MAAM,MAAM,CAAC,SAAS,EAC/E,QAAO;AAET,SAAO,EAAE,OAAO,EAAE,CAAC,aAAa,GAAG,EAAE,MAAM,EAAE;GAC7C,CACD,OAAO,QAAQ,CAEL,KAAK,IAAI,IAAI;;AAO5B,SAAS,iBAAiB,YAA8B;CACtD,MAAM,yBAAS,IAAI,KAAqB;AACxC,MAAK,MAAM,KAAK,YAAY;EAC1B,MAAM,QAAQ,EAAE,MAAM,IAAI;EAC1B,MAAM,SAAS,MAAM,SAAS,IAAI,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,IAAI,GAAG;AACjE,SAAO,IAAI,SAAS,OAAO,IAAI,OAAO,IAAI,KAAK,EAAE;;CAGnD,IAAI,OAAO;CACX,IAAI,YAAY;AAChB,MAAK,MAAM,CAAC,QAAQ,UAAU,OAC5B,KAAI,QAAQ,WAAW;AACrB,cAAY;AACZ,SAAO;;AAGX,QAAO;;;;;;;;;AAcT,SAAS,sBAAsB,YAA8B;CAC3D,MAAM,2BAAW,IAAI,KAAa;CAClC,MAAM,cAAc;AAEpB,MAAK,MAAM,KAAK,YAAY;EAC1B,MAAM,IAAI,YAAY,KAAK,EAAE;AAC7B,MAAI,EACF,UAAS,IAAI,GAAG,EAAE,GAAG,GAAG,EAAE,KAAK;OAC1B;GAEL,MAAM,YAAY,EAAE,MAAM,IAAI,CAAC;AAC/B,YAAS,IAAI,UAAU;;;AAG3B,QAAO,SAAS;;;;;;;;;;AAelB,SAAS,eACP,MACA,iBACA,eACQ;CACR,MAAM,YAAY,KAAK,IAAI,OAAO,IAAI,EAAI;CAC1C,MAAM,eAAe,KAAK,IAAI,gBAAgB,GAAG,EAAI;CACrD,MAAM,MAAM,KAAM,YAAY,MAAO,kBAAkB,MAAO;AAC9D,QAAO,KAAK,MAAM,MAAM,IAAI,GAAG;;AAOjC,eAAsB,uBACpB,SACA,QACiC;CACjC,MAAM,iBAAiB,OAAO,oBAAoB;CAClD,MAAM,WAAW,OAAO,aAAa;CACrC,MAAM,eAAe,OAAO,iBAAiB;CAC7C,MAAM,sBAAsB,OAAO,wBAAwB;CAE3D,MAAM,EAAE,YAAY,mBAAmB;AACvC,KAAI,CAAC,eACH,OAAM,IAAI,MACR,mFACD;CAIH,MAAM,cAAc,MAAM,aAAa,SAAS;EAC9C;EACA;EACA;EACA,WAAW,WAAW;EACtB;EACD,CAAC;CAEF,MAAM,QAAsB,EAAE;CAC9B,IAAI,oBAAoB;AAExB,MAAK,MAAM,SAAS,YAAY,QAAQ;EACtC,MAAM,YAAY,MAAM,MAAM,KAAK,MAAM,EAAE,KAAK;AAGhD,MAAI,MAAM,uBAAuB,SAAS,MAAM,OAAO,UAAU,EAAE;AACjE;AACA;;EAIF,MAAM,kBAAkB,iBAAiB,UAAU;EACnD,MAAM,gBAAgB,sBAAsB,UAAU;EACtD,MAAM,aAAa,eAAe,MAAM,MAAM,MAAM,iBAAiB,cAAc;EAMnF,MAAM,cAAsC,MAAM,MAAM,KAAK,GAAG,SAAS;GACvE,YAAY,EAAE;GACd,OAAO,EAAE,SAAS,EAAE,KAAK,MAAM,IAAI,CAAC,KAAK,EAAE,QAAQ,UAAU,GAAG,IAAI,EAAE;GACtE,WAAW,KAAK,OAAO,IAAO,MAAM,KAAK,IAAI,MAAM,MAAM,SAAS,GAAG,EAAE,GAAI,MAAO,IAAI,GAAG;GAC1F,EAAE;AAEH,QAAM,KAAK;GACT,IAAI,MAAM;GACV,OAAO,MAAM;GACb,MAAM,MAAM;GACZ;GACA,cAAc;GACd,iBAAiB,iBAAiB,MAAM,MAAM;GAC9C,kBAAkB;GAClB,gBAAgB;GACjB,CAAC;AAEF,MAAI,MAAM,UAAU,SAAU;;AAIhC,OAAM,MAAM,GAAG,MAAM,EAAE,aAAa,EAAE,WAAW;AAEjD,QAAO;EACL;EACA,yBAAyB,YAAY,OAAO,SAAS;EACrD,oBAAoB;EACrB;;AAOH,SAAgB,sBACd,QACA,WACuB;CACvB,MAAM,EAAE,YAAY,OAAO,QAAQ,iBAAiB;CAIpD,MAAM,WAAW,GADC,MAAM,QAAQ,iBAAiB,IAAI,CACvB;CAG9B,MAAM,YAAY,OAAO,QAAQ,cAAc,GAAG;CAClD,MAAM,aAAa,YAAY,GAAG,UAAU,GAAG,aAAa;CAG5D,MAAM,UAAU,KAAK,WAAW,WAAW;CAC3C,MAAM,SAAS,QAAQ,QAAQ;CAG/B,MAAM,YAAY,aACf,KAAK,MAAM;AAIV,SAAO,OAFM,EAAE,MAAM,IAAI,CAAC,KAAK,EAAE,QAAQ,UAAU,GAAG,IAAI,EAEvC;GACnB,CACD,KAAK,KAAK;CAEb,MAAM,gBAAgB,aAAa;CAEnC,MAAM,UAAU;EACd,KAAK;EACL;EACA,oCAAoC,WAAW;EAC/C,aAAa,cAAc;EAC3B;EACA;EACA;EACA,aAAa;EACb;EACA;EACA;EACA;EACA;EACD,CAAC,KAAK,KAAK;AAGZ,WAAU,QAAQ,EAAE,WAAW,MAAM,CAAC;AACtC,eAAc,SAAS,SAAS,QAAQ;AAExC,QAAO;EACL;EACA;EACA;EACD"}
+{"version":3,"file":"latent-ideas-DvWBRHsy.mjs","names":[],"sources":["../src/graph/latent-ideas.ts"],"sourcesContent":["/**\n * latent-ideas.ts — graph_latent_ideas and idea_materialize endpoint handlers\n *\n * \"Latent ideas\" are recurring themes in the vault that exist as embedding\n * clusters but have NO dedicated note written about them yet.  PAI surfaces\n * these by running the same agglomerative clustering used by graph_clusters /\n * zettelThemes and then filtering OUT any cluster whose label is well-matched\n * by an existing note title.\n *\n * The materialize endpoint writes a new Markdown note to the vault filesystem\n * and returns its content so the plugin can open it immediately.\n */\n\nimport { mkdirSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport { TITLE_STOP_WORDS } from \"../utils/stop-words.js\";\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport { zettelThemes } from \"../zettelkasten/themes.js\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphLatentIdeasParams {\n  project_id: number;\n  /** Minimum notes in a cluster (default: 3) */\n  min_cluster_size?: number;\n  /** Cap on returned ideas (default: 15) */\n  max_ideas?: number;\n  /** How far back to look in days (default: 180) */\n  lookback_days?: number;\n  /** Cosine similarity clustering threshold (default: 0.65) */\n  similarity_threshold?: number;\n}\n\nexport interface LatentIdeaSourceNote {\n  vault_path: string;\n  title: string;\n  /** How strongly this note relates to the theme (0-1) */\n  relevance: number;\n}\n\nexport interface LatentIdea {\n  id: number;\n  /** Auto-generated cluster label from zettelThemes */\n  label: string;\n  /** Number of notes touching this theme */\n  size: number;\n  /** 0-1, how likely this is a real coherent idea */\n  confidence: number;\n  /** Notes that contribute to this theme */\n  source_notes: LatentIdeaSourceNote[];\n  /** Cleaned-up version of label for a potential note title */\n  suggested_title: string;\n  /** Most common folder among source notes */\n  suggested_folder: string;\n  /** Number of distinct session date-folders (e.g. \"2026/03\") touching this theme */\n  sessions_count: number;\n}\n\nexport interface GraphLatentIdeasResult {\n  ideas: LatentIdea[];\n  total_clusters_analyzed: number;\n  /** How many clusters already have a matching note (excluded from results) */\n  materialized_count: number;\n}\n\n// ---------------------------------------------------------------------------\n// Materialize params / result\n// ---------------------------------------------------------------------------\n\nexport interface IdeaMaterializeParams {\n  idea_label: string;\n  /** User-chosen title for the new note */\n  title: string;\n  /** Vault-relative folder path where the note should be created */\n  folder: string;\n  /** Vault-relative paths of the source notes to link from the new note */\n  source_paths: string[];\n  project_id: number;\n}\n\nexport interface IdeaMaterializeResult {\n  /** Vault-relative path of the created note */\n  vault_path: string;\n  /** Generated markdown content */\n  content: string;\n  /** Number of wikilinks inserted */\n  links_created: number;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: check if a cluster already has a matching note\n// ---------------------------------------------------------------------------\n\n/**\n * Returns true when any existing vault note title closely matches the cluster\n * label — meaning a dedicated note already exists for this topic.\n *\n * Matching strategy (simple, fast, no embeddings needed):\n *   1. Lowercase both sides and split into words.\n *   2. Remove stop words from the label words.\n *   3. If ≥ 60% of the significant label words appear in a note title → match.\n */\n// TITLE_STOP_WORDS imported from utils/stop-words.ts\n\nfunction labelMatchesTitle(label: string, title: string): boolean {\n  const labelWords = label\n    .toLowerCase()\n    .split(/[\\s\\-_/]+/)\n    .filter((w) => w.length > 2 && !TITLE_STOP_WORDS.has(w));\n\n  if (labelWords.length === 0) return false;\n\n  const titleLower = title.toLowerCase();\n  const matchCount = labelWords.filter((w) => titleLower.includes(w)).length;\n  return matchCount / labelWords.length >= 0.6;\n}\n\n/**\n * Check whether any note indexed in the vault has a title matching the label.\n * Fetches all vault file rows via StorageBackend for efficiency.\n */\nasync function clusterHasMatchingNote(\n  backend: StorageBackend,\n  label: string,\n  notePaths: string[]\n): Promise<boolean> {\n  // First check the notes already in the cluster themselves — if any cluster\n  // member's title matches the label it IS the index note → materialized.\n  const pathSet = new Set(notePaths);\n\n  // Fetch all vault files (bounded — vault rarely > 50k notes)\n  const rows = await backend.getAllVaultFiles();\n\n  for (const row of rows) {\n    if (!row.title) continue;\n    // Skip notes already counted inside the cluster — they don't count as\n    // \"dedicated notes\"; we only skip a cluster if a SEPARATE note exists.\n    if (pathSet.has(row.vaultPath)) continue;\n    if (labelMatchesTitle(label, row.title)) return true;\n  }\n  return false;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: generate a clean suggested title\n// ---------------------------------------------------------------------------\n\nfunction toSuggestedTitle(label: string): string {\n  // Remove leading/trailing whitespace, capitalize each word, remove stop words\n  // that are all-lowercase at the start of the title.\n  const words = label\n    .trim()\n    .split(/\\s+/)\n    .map((w, i) => {\n      const lower = w.toLowerCase();\n      // Drop leading stop words (but keep if they're the only word)\n      if (i === 0 && TITLE_STOP_WORDS.has(lower) && label.trim().split(/\\s+/).length > 1) {\n        return \"\";\n      }\n      return w.charAt(0).toUpperCase() + w.slice(1);\n    })\n    .filter(Boolean);\n\n  return words.join(\" \") || label;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: find most common folder\n// ---------------------------------------------------------------------------\n\nfunction mostCommonFolder(vaultPaths: string[]): string {\n  const counts = new Map<string, number>();\n  for (const p of vaultPaths) {\n    const parts = p.split(\"/\");\n    const folder = parts.length > 1 ? parts.slice(0, -1).join(\"/\") : \"\";\n    counts.set(folder, (counts.get(folder) ?? 0) + 1);\n  }\n\n  let best = \"\";\n  let bestCount = 0;\n  for (const [folder, count] of counts) {\n    if (count > bestCount) {\n      bestCount = count;\n      best = folder;\n    }\n  }\n  return best;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: count distinct session date-folders\n// ---------------------------------------------------------------------------\n\n/**\n * Heuristic: vault notes are often stored in date-based folders like\n * \"2026/03/15\" or \"Daily/2026-03\".  We extract the first numeric path\n * segment that looks like a year (2020-2030) and group by year+month.\n *\n * Falls back to counting distinct top-level folders.\n */\nfunction countDistinctSessions(vaultPaths: string[]): number {\n  const sessions = new Set<string>();\n  const yearMonthRe = /\\b(202\\d)\\D?(0[1-9]|1[0-2])\\b/;\n\n  for (const p of vaultPaths) {\n    const m = yearMonthRe.exec(p);\n    if (m) {\n      sessions.add(`${m[1]}-${m[2]}`);\n    } else {\n      // Fallback: use top-level folder as a proxy for \"session bucket\"\n      const topFolder = p.split(\"/\")[0];\n      sessions.add(topFolder);\n    }\n  }\n  return sessions.size;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: calculate confidence score\n// ---------------------------------------------------------------------------\n\n/**\n * Confidence combines:\n *  - Cluster size (normalized, capped at 20 for max contribution)\n *  - Folder diversity (0-1 already)\n *  - Sessions count (normalized, capped at 5)\n *\n * Formula: 0.4 * sizeScore + 0.35 * folderDiversity + 0.25 * sessionScore\n */\nfunction calcConfidence(\n  size: number,\n  folderDiversity: number,\n  sessionsCount: number\n): number {\n  const sizeScore = Math.min(size / 20, 1.0);\n  const sessionScore = Math.min(sessionsCount / 5, 1.0);\n  const raw = 0.4 * sizeScore + 0.35 * folderDiversity + 0.25 * sessionScore;\n  return Math.round(raw * 100) / 100;\n}\n\n// ---------------------------------------------------------------------------\n// Main handler: graph_latent_ideas\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphLatentIdeas(\n  backend: StorageBackend,\n  params: GraphLatentIdeasParams\n): Promise<GraphLatentIdeasResult> {\n  const minClusterSize = params.min_cluster_size ?? 3;\n  const maxIdeas = params.max_ideas ?? 15;\n  const lookbackDays = params.lookback_days ?? 180;\n  const similarityThreshold = params.similarity_threshold ?? 0.65;\n\n  const { project_id: vaultProjectId } = params;\n  if (!vaultProjectId) {\n    throw new Error(\n      \"graph_latent_ideas: project_id is required (pass the vault project's numeric ID)\"\n    );\n  }\n\n  // Run the same clustering algorithm used by graph_clusters\n  const themeResult = await zettelThemes(backend, {\n    vaultProjectId,\n    lookbackDays,\n    minClusterSize,\n    maxThemes: maxIdeas * 3, // Over-fetch — many will be filtered as materialized\n    similarityThreshold,\n  });\n\n  const ideas: LatentIdea[] = [];\n  let materializedCount = 0;\n\n  for (const theme of themeResult.themes) {\n    const notePaths = theme.notes.map((n) => n.path);\n\n    // Check if a dedicated note already exists for this theme\n    if (await clusterHasMatchingNote(backend, theme.label, notePaths)) {\n      materializedCount++;\n      continue;\n    }\n\n    // This is a latent idea — no dedicated note exists yet\n    const suggestedFolder = mostCommonFolder(notePaths);\n    const sessionsCount = countDistinctSessions(notePaths);\n    const confidence = calcConfidence(theme.size, theme.folderDiversity, sessionsCount);\n\n    // Build source notes with relevance scores\n    // Relevance is approximated by position in cluster (centroid-closest first)\n    // zettelThemes returns notes in no guaranteed order; assign uniform relevance\n    // decreasing from 1.0 to 0.5 across the list.\n    const sourceNotes: LatentIdeaSourceNote[] = theme.notes.map((n, idx) => ({\n      vault_path: n.path,\n      title: n.title ?? n.path.split(\"/\").pop()?.replace(/\\.md$/i, \"\") ?? n.path,\n      relevance: Math.round((1.0 - (idx / Math.max(theme.notes.length - 1, 1)) * 0.5) * 100) / 100,\n    }));\n\n    ideas.push({\n      id: theme.id,\n      label: theme.label,\n      size: theme.size,\n      confidence,\n      source_notes: sourceNotes,\n      suggested_title: toSuggestedTitle(theme.label),\n      suggested_folder: suggestedFolder,\n      sessions_count: sessionsCount,\n    });\n\n    if (ideas.length >= maxIdeas) break;\n  }\n\n  // Sort by confidence descending\n  ideas.sort((a, b) => b.confidence - a.confidence);\n\n  return {\n    ideas,\n    total_clusters_analyzed: themeResult.themes.length + materializedCount,\n    materialized_count: materializedCount,\n  };\n}\n\n// ---------------------------------------------------------------------------\n// Materialize handler: idea_materialize\n// ---------------------------------------------------------------------------\n\nexport function handleIdeaMaterialize(\n  params: IdeaMaterializeParams,\n  vaultPath: string\n): IdeaMaterializeResult {\n  const { idea_label, title, folder, source_paths } = params;\n\n  // Sanitize filename: replace characters illegal in filenames\n  const safeTitle = title.replace(/[/\\\\:*?\"<>|]/g, \"-\");\n  const fileName = `${safeTitle}.md`;\n\n  // Vault-relative path (forward slashes, no leading slash)\n  const relFolder = folder.replace(/^\\/+|\\/+$/g, \"\");\n  const vault_path = relFolder ? `${relFolder}/${fileName}` : fileName;\n\n  // Absolute filesystem path\n  const absPath = join(vaultPath, vault_path);\n  const absDir = dirname(absPath);\n\n  // Build wikilinks from source_paths\n  const wikilinks = source_paths\n    .map((p) => {\n      // Derive a display name: filename without extension\n      const name = p.split(\"/\").pop()?.replace(/\\.md$/i, \"\") ?? p;\n      // Relative wikilink — use just the filename (Obsidian resolves by title)\n      return `- [[${name}]]`;\n    })\n    .join(\"\\n\");\n\n  const links_created = source_paths.length;\n\n  const content = [\n    `# ${title}`,\n    \"\",\n    `*Materialized from latent idea: \"${idea_label}\"*`,\n    `*Sources: ${links_created} notes*`,\n    \"\",\n    \"## Related Notes\",\n    \"\",\n    wikilinks || \"*(no source notes)*\",\n    \"\",\n    \"## Notes\",\n    \"\",\n    \"<!-- Add your thoughts about this idea here -->\",\n    \"\",\n  ].join(\"\\n\");\n\n  // Write the file (create parent directories as needed)\n  mkdirSync(absDir, { recursive: true });\n  writeFileSync(absPath, content, \"utf-8\");\n\n  return {\n    vault_path,\n    content,\n    links_created,\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AA0GA,SAAS,kBAAkB,OAAe,OAAwB;CAChE,MAAM,aAAa,MAChB,aAAa,CACb,MAAM,YAAY,CAClB,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,iBAAiB,IAAI,EAAE,CAAC;AAE1D,KAAI,WAAW,WAAW,EAAG,QAAO;CAEpC,MAAM,aAAa,MAAM,aAAa;AAEtC,QADmB,WAAW,QAAQ,MAAM,WAAW,SAAS,EAAE,CAAC,CAAC,SAChD,WAAW,UAAU;;;;;;AAO3C,eAAe,uBACb,SACA,OACA,WACkB;CAGlB,MAAM,UAAU,IAAI,IAAI,UAAU;CAGlC,MAAM,OAAO,MAAM,QAAQ,kBAAkB;AAE7C,MAAK,MAAM,OAAO,MAAM;AACtB,MAAI,CAAC,IAAI,MAAO;AAGhB,MAAI,QAAQ,IAAI,IAAI,UAAU,CAAE;AAChC,MAAI,kBAAkB,OAAO,IAAI,MAAM,CAAE,QAAO;;AAElD,QAAO;;AAOT,SAAS,iBAAiB,OAAuB;AAgB/C,QAbc,MACX,MAAM,CACN,MAAM,MAAM,CACZ,KAAK,GAAG,MAAM;EACb,MAAM,QAAQ,EAAE,aAAa;AAE7B,MAAI,MAAM,KAAK,iBAAiB,IAAI,MAAM,IAAI,MAAM,MAAM,CAAC,MAAM,MAAM,CAAC,SAAS,EAC/E,QAAO;AAET,SAAO,EAAE,OAAO,EAAE,CAAC,aAAa,GAAG,EAAE,MAAM,EAAE;GAC7C,CACD,OAAO,QAAQ,CAEL,KAAK,IAAI,IAAI;;AAO5B,SAAS,iBAAiB,YAA8B;CACtD,MAAM,yBAAS,IAAI,KAAqB;AACxC,MAAK,MAAM,KAAK,YAAY;EAC1B,MAAM,QAAQ,EAAE,MAAM,IAAI;EAC1B,MAAM,SAAS,MAAM,SAAS,IAAI,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,IAAI,GAAG;AACjE,SAAO,IAAI,SAAS,OAAO,IAAI,OAAO,IAAI,KAAK,EAAE;;CAGnD,IAAI,OAAO;CACX,IAAI,YAAY;AAChB,MAAK,MAAM,CAAC,QAAQ,UAAU,OAC5B,KAAI,QAAQ,WAAW;AACrB,cAAY;AACZ,SAAO;;AAGX,QAAO;;;;;;;;;AAcT,SAAS,sBAAsB,YAA8B;CAC3D,MAAM,2BAAW,IAAI,KAAa;CAClC,MAAM,cAAc;AAEpB,MAAK,MAAM,KAAK,YAAY;EAC1B,MAAM,IAAI,YAAY,KAAK,EAAE;AAC7B,MAAI,EACF,UAAS,IAAI,GAAG,EAAE,GAAG,GAAG,EAAE,KAAK;OAC1B;GAEL,MAAM,YAAY,EAAE,MAAM,IAAI,CAAC;AAC/B,YAAS,IAAI,UAAU;;;AAG3B,QAAO,SAAS;;;;;;;;;;AAelB,SAAS,eACP,MACA,iBACA,eACQ;CACR,MAAM,YAAY,KAAK,IAAI,OAAO,IAAI,EAAI;CAC1C,MAAM,eAAe,KAAK,IAAI,gBAAgB,GAAG,EAAI;CACrD,MAAM,MAAM,KAAM,YAAY,MAAO,kBAAkB,MAAO;AAC9D,QAAO,KAAK,MAAM,MAAM,IAAI,GAAG;;AAOjC,eAAsB,uBACpB,SACA,QACiC;CACjC,MAAM,iBAAiB,OAAO,oBAAoB;CAClD,MAAM,WAAW,OAAO,aAAa;CACrC,MAAM,eAAe,OAAO,iBAAiB;CAC7C,MAAM,sBAAsB,OAAO,wBAAwB;CAE3D,MAAM,EAAE,YAAY,mBAAmB;AACvC,KAAI,CAAC,eACH,OAAM,IAAI,MACR,mFACD;CAIH,MAAM,cAAc,MAAM,aAAa,SAAS;EAC9C;EACA;EACA;EACA,WAAW,WAAW;EACtB;EACD,CAAC;CAEF,MAAM,QAAsB,EAAE;CAC9B,IAAI,oBAAoB;AAExB,MAAK,MAAM,SAAS,YAAY,QAAQ;EACtC,MAAM,YAAY,MAAM,MAAM,KAAK,MAAM,EAAE,KAAK;AAGhD,MAAI,MAAM,uBAAuB,SAAS,MAAM,OAAO,UAAU,EAAE;AACjE;AACA;;EAIF,MAAM,kBAAkB,iBAAiB,UAAU;EACnD,MAAM,gBAAgB,sBAAsB,UAAU;EACtD,MAAM,aAAa,eAAe,MAAM,MAAM,MAAM,iBAAiB,cAAc;EAMnF,MAAM,cAAsC,MAAM,MAAM,KAAK,GAAG,SAAS;GACvE,YAAY,EAAE;GACd,OAAO,EAAE,SAAS,EAAE,KAAK,MAAM,IAAI,CAAC,KAAK,EAAE,QAAQ,UAAU,GAAG,IAAI,EAAE;GACtE,WAAW,KAAK,OAAO,IAAO,MAAM,KAAK,IAAI,MAAM,MAAM,SAAS,GAAG,EAAE,GAAI,MAAO,IAAI,GAAG;GAC1F,EAAE;AAEH,QAAM,KAAK;GACT,IAAI,MAAM;GACV,OAAO,MAAM;GACb,MAAM,MAAM;GACZ;GACA,cAAc;GACd,iBAAiB,iBAAiB,MAAM,MAAM;GAC9C,kBAAkB;GAClB,gBAAgB;GACjB,CAAC;AAEF,MAAI,MAAM,UAAU,SAAU;;AAIhC,OAAM,MAAM,GAAG,MAAM,EAAE,aAAa,EAAE,WAAW;AAEjD,QAAO;EACL;EACA,yBAAyB,YAAY,OAAO,SAAS;EACrD,oBAAoB;EACrB;;AAOH,SAAgB,sBACd,QACA,WACuB;CACvB,MAAM,EAAE,YAAY,OAAO,QAAQ,iBAAiB;CAIpD,MAAM,WAAW,GADC,MAAM,QAAQ,iBAAiB,IAAI,CACvB;CAG9B,MAAM,YAAY,OAAO,QAAQ,cAAc,GAAG;CAClD,MAAM,aAAa,YAAY,GAAG,UAAU,GAAG,aAAa;CAG5D,MAAM,UAAU,KAAK,WAAW,WAAW;CAC3C,MAAM,SAAS,QAAQ,QAAQ;CAG/B,MAAM,YAAY,aACf,KAAK,MAAM;AAIV,SAAO,OAFM,EAAE,MAAM,IAAI,CAAC,KAAK,EAAE,QAAQ,UAAU,GAAG,IAAI,EAEvC;GACnB,CACD,KAAK,KAAK;CAEb,MAAM,gBAAgB,aAAa;CAEnC,MAAM,UAAU;EACd,KAAK;EACL;EACA,oCAAoC,WAAW;EAC/C,aAAa,cAAc;EAC3B;EACA;EACA;EACA,aAAa;EACb;EACA;EACA;EACA;EACA;EACD,CAAC,KAAK,KAAK;AAGZ,WAAU,QAAQ,EAAE,WAAW,MAAM,CAAC;AACtC,eAAc,SAAS,SAAS,QAAQ;AAExC,QAAO;EACL;EACA;EACA;EACD"}

package/dist/{neighborhood-BYYbEkUJ.mjs → neighborhood-u8ytjmWq.mjs}

@@ -132,4 +132,4 @@ async function handleGraphNeighborhood(pool, backend, params) {
 
 //#endregion
 export { handleGraphNeighborhood };
-//# sourceMappingURL=neighborhood-BYYbEkUJ.mjs.map
+//# sourceMappingURL=neighborhood-u8ytjmWq.mjs.map

package/dist/{neighborhood-BYYbEkUJ.mjs.map → neighborhood-u8ytjmWq.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"neighborhood-BYYbEkUJ.mjs","names":[],"sources":["../src/graph/neighborhood.ts"],"sourcesContent":["/**\n * neighborhood.ts — graph_neighborhood endpoint handler\n *\n * Given a set of vault note paths (typically the notes inside a cluster),\n * returns the individual note nodes and the wikilink edges between them.\n *\n * Optionally enriches with semantic edges computed from cosine similarity\n * between chunk embeddings stored in the federation database.\n */\n\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport type { Pool } from \"pg\";\nimport { deserializeEmbedding } from \"../memory/embeddings.js\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphNeighborhoodParams {\n  /** Vault-relative paths of notes in the cluster */\n  vault_paths: string[];\n  /** Numeric PAI project ID */\n  project_id: number;\n  /** Whether to compute semantic similarity edges (default: false) */\n  include_semantic_edges?: boolean;\n  /** Cosine similarity threshold for semantic edges (default: 0.7) */\n  semantic_threshold?: number;\n}\n\nexport interface NoteNode {\n  vault_path: string;\n  title: string;\n  folder: string;\n  observation_types: Record<string, number>;\n  dominant_type: string;\n  updated_at: number;\n  word_count: number;\n}\n\nexport interface NoteEdge {\n  source: string;\n  target: string;\n  type: \"wikilink\" | \"semantic\";\n  weight: number;\n}\n\nexport interface GraphNeighborhoodResult {\n  nodes: NoteNode[];\n  edges: NoteEdge[];\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\nfunction folderFromPath(vaultPath: string): string {\n  const lastSlash = vaultPath.lastIndexOf(\"/\");\n  return lastSlash === -1 ? \"\" : vaultPath.slice(0, lastSlash);\n}\n\nfunction cosineSimilarity(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length === 0) return 0;\n  let dot = 0;\n  let normA = 0;\n  let normB = 0;\n  for (let i = 0; i < a.length; i++) {\n    dot += a[i] * b[i];\n    normA += a[i] * a[i];\n    normB += b[i] * b[i];\n  }\n  if (normA === 0 || normB === 0) return 0;\n  return dot / (Math.sqrt(normA) * Math.sqrt(normB));\n}\n\nfunction dominantType(counts: Record<string, number>): string {\n  let dominant = \"unknown\";\n  let maxCount = 0;\n  for (const [type, n] of Object.entries(counts)) {\n    if (n > maxCount) {\n      maxCount = n;\n      dominant = type;\n    }\n  }\n  return dominant;\n}\n\n// ---------------------------------------------------------------------------\n// Observation type enrichment (same pattern as clusters.ts)\n// ---------------------------------------------------------------------------\n\nasync function fetchObservationTypes(\n  pool: Pool,\n  filePaths: string[],\n  projectId: number\n): Promise<Map<string, Record<string, number>>> {\n  if (filePaths.length === 0) return new Map();\n\n  try {\n    const params: (string[] | number)[] = [filePaths, projectId];\n\n    const result = await pool.query<{ path: string; type: string; cnt: string }>(\n      `SELECT unnested_path AS path, type, COUNT(*) AS cnt\n       FROM pai_observations,\n            LATERAL unnest(files_modified || files_read) AS unnested_path\n       WHERE unnested_path = ANY($1::text[])\n         AND project_id = $2\n       GROUP BY unnested_path, type`,\n      params\n    );\n\n    const byPath = new Map<string, Record<string, number>>();\n    for (const row of result.rows) {\n      const existing = byPath.get(row.path) ?? {};\n      existing[row.type] = (existing[row.type] ?? 0) + parseInt(row.cnt, 10);\n      byPath.set(row.path, existing);\n    }\n    return byPath;\n  } catch {\n    return new Map();\n  }\n}\n\n// ---------------------------------------------------------------------------\n// Main handler\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphNeighborhood(\n  pool: Pool | null,\n  backend: StorageBackend,\n  params: GraphNeighborhoodParams\n): Promise<GraphNeighborhoodResult> {\n  const vaultPaths = params.vault_paths ?? [];\n  if (vaultPaths.length === 0) {\n    return { nodes: [], edges: [] };\n  }\n\n  const includeSemanticEdges = params.include_semantic_edges ?? false;\n  const semanticThreshold = params.semantic_threshold ?? 0.7;\n\n  // -------------------------------------------------------------------------\n  // 1. Fetch node metadata from vault_files\n  // -------------------------------------------------------------------------\n\n  const fileRows = await backend.getVaultFilesByPaths(vaultPaths);\n\n  const fileIndex = new Map<string, { vaultPath: string; title: string | null; indexedAt: number }>();\n  for (const row of fileRows) {\n    fileIndex.set(row.vaultPath, row);\n  }\n\n  // -------------------------------------------------------------------------\n  // 2. Fetch observation types (Postgres if available)\n  // -------------------------------------------------------------------------\n\n  const observationsByPath =\n    pool !== null\n      ? await fetchObservationTypes(pool, vaultPaths, params.project_id)\n      : new Map<string, Record<string, number>>();\n\n  // -------------------------------------------------------------------------\n  // 3. Build NoteNode array\n  // -------------------------------------------------------------------------\n\n  const nodes: NoteNode[] = vaultPaths.map((vp) => {\n    const fileRow = fileIndex.get(vp);\n    const fileName = vp.split(\"/\").pop() ?? vp;\n    const rawTitle = fileRow?.title ?? fileName.replace(/\\.md$/i, \"\");\n\n    const obsCounts = observationsByPath.get(vp) ?? {};\n\n    return {\n      vault_path: vp,\n      title: rawTitle,\n      folder: folderFromPath(vp),\n      observation_types: obsCounts,\n      dominant_type: dominantType(obsCounts),\n      updated_at: fileRow?.indexedAt ?? 0,\n      word_count: 0,\n    };\n  });\n\n  // -------------------------------------------------------------------------\n  // 4. Fetch wikilink edges between the provided paths\n  // -------------------------------------------------------------------------\n\n  const pathSet = new Set(vaultPaths);\n  const linkRows = await backend.getVaultLinksFromPaths(vaultPaths);\n\n  const edges: NoteEdge[] = [];\n\n  for (const row of linkRows) {\n    if (!row.targetPath || !pathSet.has(row.targetPath)) continue;\n\n    edges.push({\n      source: row.sourcePath,\n      target: row.targetPath,\n      type: \"wikilink\",\n      weight: 1.0,\n    });\n  }\n\n  // -------------------------------------------------------------------------\n  // 5. Optional: semantic edges\n  // -------------------------------------------------------------------------\n\n  if (includeSemanticEdges && vaultPaths.length > 1) {\n    // Fetch mean embeddings for all paths\n    const embeddings = new Map<string, number[]>();\n    for (const vp of vaultPaths) {\n      const chunkRows = await backend.getChunksForPath(params.project_id, vp);\n      const embRows = chunkRows.filter(r => r.embedding !== null) as Array<{ text: string; embedding: Buffer }>;\n      if (embRows.length === 0) continue;\n\n      let vecLen = 0;\n      const vectors: Float32Array[] = [];\n\n      for (const row of embRows) {\n        const arr = deserializeEmbedding(row.embedding);\n        if (vecLen === 0) vecLen = arr.length;\n        if (arr.length === vecLen) vectors.push(arr);\n      }\n\n      if (vectors.length === 0 || vecLen === 0) continue;\n\n      const mean = new Array<number>(vecLen).fill(0);\n      for (const vec of vectors) {\n        for (let i = 0; i < vecLen; i++) {\n          mean[i] += vec[i];\n        }\n      }\n      for (let i = 0; i < vecLen; i++) {\n        mean[i] /= vectors.length;\n      }\n      embeddings.set(vp, mean);\n    }\n\n    const existingEdgeKeys = new Set<string>(\n      edges.map((e) => `${e.source}|||${e.target}`)\n    );\n\n    const pathsWithEmbeddings = Array.from(embeddings.keys());\n    for (let i = 0; i < pathsWithEmbeddings.length; i++) {\n      for (let j = i + 1; j < pathsWithEmbeddings.length; j++) {\n        const pathA = pathsWithEmbeddings[i];\n        const pathB = pathsWithEmbeddings[j];\n\n        const vecA = embeddings.get(pathA)!;\n        const vecB = embeddings.get(pathB)!;\n\n        const sim = cosineSimilarity(vecA, vecB);\n        if (sim < semanticThreshold) continue;\n\n        const keyAB = `${pathA}|||${pathB}`;\n        const keyBA = `${pathB}|||${pathA}`;\n        if (existingEdgeKeys.has(keyAB) || existingEdgeKeys.has(keyBA)) continue;\n\n        edges.push({\n          source: pathA,\n          target: pathB,\n          type: \"semantic\",\n          weight: sim,\n        });\n        existingEdgeKeys.add(keyAB);\n      }\n    }\n  }\n\n  return { nodes, edges };\n}\n"],"mappings":";;;AAuDA,SAAS,eAAe,WAA2B;CACjD,MAAM,YAAY,UAAU,YAAY,IAAI;AAC5C,QAAO,cAAc,KAAK,KAAK,UAAU,MAAM,GAAG,UAAU;;AAG9D,SAAS,iBAAiB,GAAa,GAAqB;AAC1D,KAAI,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAG,QAAO;CACpD,IAAI,MAAM;CACV,IAAI,QAAQ;CACZ,IAAI,QAAQ;AACZ,MAAK,IAAI,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,SAAO,EAAE,KAAK,EAAE;AAChB,WAAS,EAAE,KAAK,EAAE;AAClB,WAAS,EAAE,KAAK,EAAE;;AAEpB,KAAI,UAAU,KAAK,UAAU,EAAG,QAAO;AACvC,QAAO,OAAO,KAAK,KAAK,MAAM,GAAG,KAAK,KAAK,MAAM;;AAGnD,SAAS,aAAa,QAAwC;CAC5D,IAAI,WAAW;CACf,IAAI,WAAW;AACf,MAAK,MAAM,CAAC,MAAM,MAAM,OAAO,QAAQ,OAAO,CAC5C,KAAI,IAAI,UAAU;AAChB,aAAW;AACX,aAAW;;AAGf,QAAO;;AAOT,eAAe,sBACb,MACA,WACA,WAC8C;AAC9C,KAAI,UAAU,WAAW,EAAG,wBAAO,IAAI,KAAK;AAE5C,KAAI;EACF,MAAM,SAAgC,CAAC,WAAW,UAAU;EAE5D,MAAM,SAAS,MAAM,KAAK,MACxB;;;;;sCAMA,OACD;EAED,MAAM,yBAAS,IAAI,KAAqC;AACxD,OAAK,MAAM,OAAO,OAAO,MAAM;GAC7B,MAAM,WAAW,OAAO,IAAI,IAAI,KAAK,IAAI,EAAE;AAC3C,YAAS,IAAI,SAAS,SAAS,IAAI,SAAS,KAAK,SAAS,IAAI,KAAK,GAAG;AACtE,UAAO,IAAI,IAAI,MAAM,SAAS;;AAEhC,SAAO;SACD;AACN,yBAAO,IAAI,KAAK;;;AAQpB,eAAsB,wBACpB,MACA,SACA,QACkC;CAClC,MAAM,aAAa,OAAO,eAAe,EAAE;AAC3C,KAAI,WAAW,WAAW,EACxB,QAAO;EAAE,OAAO,EAAE;EAAE,OAAO,EAAE;EAAE;CAGjC,MAAM,uBAAuB,OAAO,0BAA0B;CAC9D,MAAM,oBAAoB,OAAO,sBAAsB;CAMvD,MAAM,WAAW,MAAM,QAAQ,qBAAqB,WAAW;CAE/D,MAAM,4BAAY,IAAI,KAA6E;AACnG,MAAK,MAAM,OAAO,SAChB,WAAU,IAAI,IAAI,WAAW,IAAI;CAOnC,MAAM,qBACJ,SAAS,OACL,MAAM,sBAAsB,MAAM,YAAY,OAAO,WAAW,mBAChE,IAAI,KAAqC;CAM/C,MAAM,QAAoB,WAAW,KAAK,OAAO;EAC/C,MAAM,UAAU,UAAU,IAAI,GAAG;EACjC,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,KAAK,IAAI;EACxC,MAAM,WAAW,SAAS,SAAS,SAAS,QAAQ,UAAU,GAAG;EAEjE,MAAM,YAAY,mBAAmB,IAAI,GAAG,IAAI,EAAE;AAElD,SAAO;GACL,YAAY;GACZ,OAAO;GACP,QAAQ,eAAe,GAAG;GAC1B,mBAAmB;GACnB,eAAe,aAAa,UAAU;GACtC,YAAY,SAAS,aAAa;GAClC,YAAY;GACb;GACD;CAMF,MAAM,UAAU,IAAI,IAAI,WAAW;CACnC,MAAM,WAAW,MAAM,QAAQ,uBAAuB,WAAW;CAEjE,MAAM,QAAoB,EAAE;AAE5B,MAAK,MAAM,OAAO,UAAU;AAC1B,MAAI,CAAC,IAAI,cAAc,CAAC,QAAQ,IAAI,IAAI,WAAW,CAAE;AAErD,QAAM,KAAK;GACT,QAAQ,IAAI;GACZ,QAAQ,IAAI;GACZ,MAAM;GACN,QAAQ;GACT,CAAC;;AAOJ,KAAI,wBAAwB,WAAW,SAAS,GAAG;EAEjD,MAAM,6BAAa,IAAI,KAAuB;AAC9C,OAAK,MAAM,MAAM,YAAY;GAE3B,MAAM,WADY,MAAM,QAAQ,iBAAiB,OAAO,YAAY,GAAG,EAC7C,QAAO,MAAK,EAAE,cAAc,KAAK;AAC3D,OAAI,QAAQ,WAAW,EAAG;GAE1B,IAAI,SAAS;GACb,MAAM,UAA0B,EAAE;AAElC,QAAK,MAAM,OAAO,SAAS;IACzB,MAAM,MAAM,qBAAqB,IAAI,UAAU;AAC/C,QAAI,WAAW,EAAG,UAAS,IAAI;AAC/B,QAAI,IAAI,WAAW,OAAQ,SAAQ,KAAK,IAAI;;AAG9C,OAAI,QAAQ,WAAW,KAAK,WAAW,EAAG;GAE1C,MAAM,OAAO,IAAI,MAAc,OAAO,CAAC,KAAK,EAAE;AAC9C,QAAK,MAAM,OAAO,QAChB,MAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,IAC1B,MAAK,MAAM,IAAI;AAGnB,QAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,IAC1B,MAAK,MAAM,QAAQ;AAErB,cAAW,IAAI,IAAI,KAAK;;EAG1B,MAAM,mBAAmB,IAAI,IAC3B,MAAM,KAAK,MAAM,GAAG,EAAE,OAAO,KAAK,EAAE,SAAS,CAC9C;EAED,MAAM,sBAAsB,MAAM,KAAK,WAAW,MAAM,CAAC;AACzD,OAAK,IAAI,IAAI,GAAG,IAAI,oBAAoB,QAAQ,IAC9C,MAAK,IAAI,IAAI,IAAI,GAAG,IAAI,oBAAoB,QAAQ,KAAK;GACvD,MAAM,QAAQ,oBAAoB;GAClC,MAAM,QAAQ,oBAAoB;GAKlC,MAAM,MAAM,iBAHC,WAAW,IAAI,MAAM,EACrB,WAAW,IAAI,MAAM,CAEM;AACxC,OAAI,MAAM,kBAAmB;GAE7B,MAAM,QAAQ,GAAG,MAAM,KAAK;GAC5B,MAAM,QAAQ,GAAG,MAAM,KAAK;AAC5B,OAAI,iBAAiB,IAAI,MAAM,IAAI,iBAAiB,IAAI,MAAM,CAAE;AAEhE,SAAM,KAAK;IACT,QAAQ;IACR,QAAQ;IACR,MAAM;IACN,QAAQ;IACT,CAAC;AACF,oBAAiB,IAAI,MAAM;;;AAKjC,QAAO;EAAE;EAAO;EAAO"}
+{"version":3,"file":"neighborhood-u8ytjmWq.mjs","names":[],"sources":["../src/graph/neighborhood.ts"],"sourcesContent":["/**\n * neighborhood.ts — graph_neighborhood endpoint handler\n *\n * Given a set of vault note paths (typically the notes inside a cluster),\n * returns the individual note nodes and the wikilink edges between them.\n *\n * Optionally enriches with semantic edges computed from cosine similarity\n * between chunk embeddings stored in the federation database.\n */\n\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport type { Pool } from \"pg\";\nimport { deserializeEmbedding } from \"../memory/embeddings.js\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphNeighborhoodParams {\n  /** Vault-relative paths of notes in the cluster */\n  vault_paths: string[];\n  /** Numeric PAI project ID */\n  project_id: number;\n  /** Whether to compute semantic similarity edges (default: false) */\n  include_semantic_edges?: boolean;\n  /** Cosine similarity threshold for semantic edges (default: 0.7) */\n  semantic_threshold?: number;\n}\n\nexport interface NoteNode {\n  vault_path: string;\n  title: string;\n  folder: string;\n  observation_types: Record<string, number>;\n  dominant_type: string;\n  updated_at: number;\n  word_count: number;\n}\n\nexport interface NoteEdge {\n  source: string;\n  target: string;\n  type: \"wikilink\" | \"semantic\";\n  weight: number;\n}\n\nexport interface GraphNeighborhoodResult {\n  nodes: NoteNode[];\n  edges: NoteEdge[];\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\nfunction folderFromPath(vaultPath: string): string {\n  const lastSlash = vaultPath.lastIndexOf(\"/\");\n  return lastSlash === -1 ? \"\" : vaultPath.slice(0, lastSlash);\n}\n\nfunction cosineSimilarity(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length === 0) return 0;\n  let dot = 0;\n  let normA = 0;\n  let normB = 0;\n  for (let i = 0; i < a.length; i++) {\n    dot += a[i] * b[i];\n    normA += a[i] * a[i];\n    normB += b[i] * b[i];\n  }\n  if (normA === 0 || normB === 0) return 0;\n  return dot / (Math.sqrt(normA) * Math.sqrt(normB));\n}\n\nfunction dominantType(counts: Record<string, number>): string {\n  let dominant = \"unknown\";\n  let maxCount = 0;\n  for (const [type, n] of Object.entries(counts)) {\n    if (n > maxCount) {\n      maxCount = n;\n      dominant = type;\n    }\n  }\n  return dominant;\n}\n\n// ---------------------------------------------------------------------------\n// Observation type enrichment (same pattern as clusters.ts)\n// ---------------------------------------------------------------------------\n\nasync function fetchObservationTypes(\n  pool: Pool,\n  filePaths: string[],\n  projectId: number\n): Promise<Map<string, Record<string, number>>> {\n  if (filePaths.length === 0) return new Map();\n\n  try {\n    const params: (string[] | number)[] = [filePaths, projectId];\n\n    const result = await pool.query<{ path: string; type: string; cnt: string }>(\n      `SELECT unnested_path AS path, type, COUNT(*) AS cnt\n       FROM pai_observations,\n            LATERAL unnest(files_modified || files_read) AS unnested_path\n       WHERE unnested_path = ANY($1::text[])\n         AND project_id = $2\n       GROUP BY unnested_path, type`,\n      params\n    );\n\n    const byPath = new Map<string, Record<string, number>>();\n    for (const row of result.rows) {\n      const existing = byPath.get(row.path) ?? {};\n      existing[row.type] = (existing[row.type] ?? 0) + parseInt(row.cnt, 10);\n      byPath.set(row.path, existing);\n    }\n    return byPath;\n  } catch {\n    return new Map();\n  }\n}\n\n// ---------------------------------------------------------------------------\n// Main handler\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphNeighborhood(\n  pool: Pool | null,\n  backend: StorageBackend,\n  params: GraphNeighborhoodParams\n): Promise<GraphNeighborhoodResult> {\n  const vaultPaths = params.vault_paths ?? [];\n  if (vaultPaths.length === 0) {\n    return { nodes: [], edges: [] };\n  }\n\n  const includeSemanticEdges = params.include_semantic_edges ?? false;\n  const semanticThreshold = params.semantic_threshold ?? 0.7;\n\n  // -------------------------------------------------------------------------\n  // 1. Fetch node metadata from vault_files\n  // -------------------------------------------------------------------------\n\n  const fileRows = await backend.getVaultFilesByPaths(vaultPaths);\n\n  const fileIndex = new Map<string, { vaultPath: string; title: string | null; indexedAt: number }>();\n  for (const row of fileRows) {\n    fileIndex.set(row.vaultPath, row);\n  }\n\n  // -------------------------------------------------------------------------\n  // 2. Fetch observation types (Postgres if available)\n  // -------------------------------------------------------------------------\n\n  const observationsByPath =\n    pool !== null\n      ? await fetchObservationTypes(pool, vaultPaths, params.project_id)\n      : new Map<string, Record<string, number>>();\n\n  // -------------------------------------------------------------------------\n  // 3. Build NoteNode array\n  // -------------------------------------------------------------------------\n\n  const nodes: NoteNode[] = vaultPaths.map((vp) => {\n    const fileRow = fileIndex.get(vp);\n    const fileName = vp.split(\"/\").pop() ?? vp;\n    const rawTitle = fileRow?.title ?? fileName.replace(/\\.md$/i, \"\");\n\n    const obsCounts = observationsByPath.get(vp) ?? {};\n\n    return {\n      vault_path: vp,\n      title: rawTitle,\n      folder: folderFromPath(vp),\n      observation_types: obsCounts,\n      dominant_type: dominantType(obsCounts),\n      updated_at: fileRow?.indexedAt ?? 0,\n      word_count: 0,\n    };\n  });\n\n  // -------------------------------------------------------------------------\n  // 4. Fetch wikilink edges between the provided paths\n  // -------------------------------------------------------------------------\n\n  const pathSet = new Set(vaultPaths);\n  const linkRows = await backend.getVaultLinksFromPaths(vaultPaths);\n\n  const edges: NoteEdge[] = [];\n\n  for (const row of linkRows) {\n    if (!row.targetPath || !pathSet.has(row.targetPath)) continue;\n\n    edges.push({\n      source: row.sourcePath,\n      target: row.targetPath,\n      type: \"wikilink\",\n      weight: 1.0,\n    });\n  }\n\n  // -------------------------------------------------------------------------\n  // 5. Optional: semantic edges\n  // -------------------------------------------------------------------------\n\n  if (includeSemanticEdges && vaultPaths.length > 1) {\n    // Fetch mean embeddings for all paths\n    const embeddings = new Map<string, number[]>();\n    for (const vp of vaultPaths) {\n      const chunkRows = await backend.getChunksForPath(params.project_id, vp);\n      const embRows = chunkRows.filter(r => r.embedding !== null) as Array<{ text: string; embedding: Buffer }>;\n      if (embRows.length === 0) continue;\n\n      let vecLen = 0;\n      const vectors: Float32Array[] = [];\n\n      for (const row of embRows) {\n        const arr = deserializeEmbedding(row.embedding);\n        if (vecLen === 0) vecLen = arr.length;\n        if (arr.length === vecLen) vectors.push(arr);\n      }\n\n      if (vectors.length === 0 || vecLen === 0) continue;\n\n      const mean = new Array<number>(vecLen).fill(0);\n      for (const vec of vectors) {\n        for (let i = 0; i < vecLen; i++) {\n          mean[i] += vec[i];\n        }\n      }\n      for (let i = 0; i < vecLen; i++) {\n        mean[i] /= vectors.length;\n      }\n      embeddings.set(vp, mean);\n    }\n\n    const existingEdgeKeys = new Set<string>(\n      edges.map((e) => `${e.source}|||${e.target}`)\n    );\n\n    const pathsWithEmbeddings = Array.from(embeddings.keys());\n    for (let i = 0; i < pathsWithEmbeddings.length; i++) {\n      for (let j = i + 1; j < pathsWithEmbeddings.length; j++) {\n        const pathA = pathsWithEmbeddings[i];\n        const pathB = pathsWithEmbeddings[j];\n\n        const vecA = embeddings.get(pathA)!;\n        const vecB = embeddings.get(pathB)!;\n\n        const sim = cosineSimilarity(vecA, vecB);\n        if (sim < semanticThreshold) continue;\n\n        const keyAB = `${pathA}|||${pathB}`;\n        const keyBA = `${pathB}|||${pathA}`;\n        if (existingEdgeKeys.has(keyAB) || existingEdgeKeys.has(keyBA)) continue;\n\n        edges.push({\n          source: pathA,\n          target: pathB,\n          type: \"semantic\",\n          weight: sim,\n        });\n        existingEdgeKeys.add(keyAB);\n      }\n    }\n  }\n\n  return { nodes, edges };\n}\n"],"mappings":";;;AAuDA,SAAS,eAAe,WAA2B;CACjD,MAAM,YAAY,UAAU,YAAY,IAAI;AAC5C,QAAO,cAAc,KAAK,KAAK,UAAU,MAAM,GAAG,UAAU;;AAG9D,SAAS,iBAAiB,GAAa,GAAqB;AAC1D,KAAI,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAG,QAAO;CACpD,IAAI,MAAM;CACV,IAAI,QAAQ;CACZ,IAAI,QAAQ;AACZ,MAAK,IAAI,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,SAAO,EAAE,KAAK,EAAE;AAChB,WAAS,EAAE,KAAK,EAAE;AAClB,WAAS,EAAE,KAAK,EAAE;;AAEpB,KAAI,UAAU,KAAK,UAAU,EAAG,QAAO;AACvC,QAAO,OAAO,KAAK,KAAK,MAAM,GAAG,KAAK,KAAK,MAAM;;AAGnD,SAAS,aAAa,QAAwC;CAC5D,IAAI,WAAW;CACf,IAAI,WAAW;AACf,MAAK,MAAM,CAAC,MAAM,MAAM,OAAO,QAAQ,OAAO,CAC5C,KAAI,IAAI,UAAU;AAChB,aAAW;AACX,aAAW;;AAGf,QAAO;;AAOT,eAAe,sBACb,MACA,WACA,WAC8C;AAC9C,KAAI,UAAU,WAAW,EAAG,wBAAO,IAAI,KAAK;AAE5C,KAAI;EACF,MAAM,SAAgC,CAAC,WAAW,UAAU;EAE5D,MAAM,SAAS,MAAM,KAAK,MACxB;;;;;sCAMA,OACD;EAED,MAAM,yBAAS,IAAI,KAAqC;AACxD,OAAK,MAAM,OAAO,OAAO,MAAM;GAC7B,MAAM,WAAW,OAAO,IAAI,IAAI,KAAK,IAAI,EAAE;AAC3C,YAAS,IAAI,SAAS,SAAS,IAAI,SAAS,KAAK,SAAS,IAAI,KAAK,GAAG;AACtE,UAAO,IAAI,IAAI,MAAM,SAAS;;AAEhC,SAAO;SACD;AACN,yBAAO,IAAI,KAAK;;;AAQpB,eAAsB,wBACpB,MACA,SACA,QACkC;CAClC,MAAM,aAAa,OAAO,eAAe,EAAE;AAC3C,KAAI,WAAW,WAAW,EACxB,QAAO;EAAE,OAAO,EAAE;EAAE,OAAO,EAAE;EAAE;CAGjC,MAAM,uBAAuB,OAAO,0BAA0B;CAC9D,MAAM,oBAAoB,OAAO,sBAAsB;CAMvD,MAAM,WAAW,MAAM,QAAQ,qBAAqB,WAAW;CAE/D,MAAM,4BAAY,IAAI,KAA6E;AACnG,MAAK,MAAM,OAAO,SAChB,WAAU,IAAI,IAAI,WAAW,IAAI;CAOnC,MAAM,qBACJ,SAAS,OACL,MAAM,sBAAsB,MAAM,YAAY,OAAO,WAAW,mBAChE,IAAI,KAAqC;CAM/C,MAAM,QAAoB,WAAW,KAAK,OAAO;EAC/C,MAAM,UAAU,UAAU,IAAI,GAAG;EACjC,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,KAAK,IAAI;EACxC,MAAM,WAAW,SAAS,SAAS,SAAS,QAAQ,UAAU,GAAG;EAEjE,MAAM,YAAY,mBAAmB,IAAI,GAAG,IAAI,EAAE;AAElD,SAAO;GACL,YAAY;GACZ,OAAO;GACP,QAAQ,eAAe,GAAG;GAC1B,mBAAmB;GACnB,eAAe,aAAa,UAAU;GACtC,YAAY,SAAS,aAAa;GAClC,YAAY;GACb;GACD;CAMF,MAAM,UAAU,IAAI,IAAI,WAAW;CACnC,MAAM,WAAW,MAAM,QAAQ,uBAAuB,WAAW;CAEjE,MAAM,QAAoB,EAAE;AAE5B,MAAK,MAAM,OAAO,UAAU;AAC1B,MAAI,CAAC,IAAI,cAAc,CAAC,QAAQ,IAAI,IAAI,WAAW,CAAE;AAErD,QAAM,KAAK;GACT,QAAQ,IAAI;GACZ,QAAQ,IAAI;GACZ,MAAM;GACN,QAAQ;GACT,CAAC;;AAOJ,KAAI,wBAAwB,WAAW,SAAS,GAAG;EAEjD,MAAM,6BAAa,IAAI,KAAuB;AAC9C,OAAK,MAAM,MAAM,YAAY;GAE3B,MAAM,WADY,MAAM,QAAQ,iBAAiB,OAAO,YAAY,GAAG,EAC7C,QAAO,MAAK,EAAE,cAAc,KAAK;AAC3D,OAAI,QAAQ,WAAW,EAAG;GAE1B,IAAI,SAAS;GACb,MAAM,UAA0B,EAAE;AAElC,QAAK,MAAM,OAAO,SAAS;IACzB,MAAM,MAAM,qBAAqB,IAAI,UAAU;AAC/C,QAAI,WAAW,EAAG,UAAS,IAAI;AAC/B,QAAI,IAAI,WAAW,OAAQ,SAAQ,KAAK,IAAI;;AAG9C,OAAI,QAAQ,WAAW,KAAK,WAAW,EAAG;GAE1C,MAAM,OAAO,IAAI,MAAc,OAAO,CAAC,KAAK,EAAE;AAC9C,QAAK,MAAM,OAAO,QAChB,MAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,IAC1B,MAAK,MAAM,IAAI;AAGnB,QAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,IAC1B,MAAK,MAAM,QAAQ;AAErB,cAAW,IAAI,IAAI,KAAK;;EAG1B,MAAM,mBAAmB,IAAI,IAC3B,MAAM,KAAK,MAAM,GAAG,EAAE,OAAO,KAAK,EAAE,SAAS,CAC9C;EAED,MAAM,sBAAsB,MAAM,KAAK,WAAW,MAAM,CAAC;AACzD,OAAK,IAAI,IAAI,GAAG,IAAI,oBAAoB,QAAQ,IAC9C,MAAK,IAAI,IAAI,IAAI,GAAG,IAAI,oBAAoB,QAAQ,KAAK;GACvD,MAAM,QAAQ,oBAAoB;GAClC,MAAM,QAAQ,oBAAoB;GAKlC,MAAM,MAAM,iBAHC,WAAW,IAAI,MAAM,EACrB,WAAW,IAAI,MAAM,CAEM;AACxC,OAAI,MAAM,kBAAmB;GAE7B,MAAM,QAAQ,GAAG,MAAM,KAAK;GAC5B,MAAM,QAAQ,GAAG,MAAM,KAAK;AAC5B,OAAI,iBAAiB,IAAI,MAAM,IAAI,iBAAiB,IAAI,MAAM,CAAE;AAEhE,SAAM,KAAK;IACT,QAAQ;IACR,QAAQ;IACR,MAAM;IACN,QAAQ;IACT,CAAC;AACF,oBAAiB,IAAI,MAAM;;;AAKjC,QAAO;EAAE;EAAO;EAAO"}

package/dist/{note-context-BK24bX8Y.mjs → note-context-CG2_e-0W.mjs}

@@ -123,4 +123,4 @@ async function handleGraphNoteContext(pool, backend, params) {
 
 //#endregion
 export { handleGraphNoteContext };
-//# sourceMappingURL=note-context-BK24bX8Y.mjs.map
+//# sourceMappingURL=note-context-CG2_e-0W.mjs.map