@tekmidian/pai 0.5.1 → 0.5.2

package/dist/{indexer-CKQcgKsz.mjs.map → indexer-CMPOiY1r.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"indexer-CKQcgKsz.mjs","names":[],"sources":["../src/memory/indexer.ts"],"sourcesContent":["/**\n * File indexer for the PAI federation memory engine.\n *\n * Scans project memory/ and Notes/ directories, chunks markdown files, and\n * inserts the resulting chunks into federation.db for BM25 search.\n *\n * Change detection: files whose SHA-256 hash has not changed since the last\n * index run are skipped, keeping incremental re-indexing fast.\n *\n * Phase 2.5: adds embedChunks() for generating vector embeddings on indexed\n * chunks that do not yet have an embedding stored.\n */\n\nimport { createHash } from \"node:crypto\";\nimport { readFileSync, statSync, readdirSync, existsSync } from \"node:fs\";\nimport { join, relative, basename, normalize } from \"node:path\";\nimport { homedir } from \"node:os\";\nimport type { Database } from \"better-sqlite3\";\nimport { chunkMarkdown } from \"./chunker.js\";\n\n// ---------------------------------------------------------------------------\n// Types\n// ---------------------------------------------------------------------------\n\nexport interface IndexResult {\n  filesProcessed: number;\n  chunksCreated: number;\n  filesSkipped: number;\n}\n\n// ---------------------------------------------------------------------------\n// Tier detection\n// ---------------------------------------------------------------------------\n\n/**\n * Classify a relative file path into one of the four memory tiers.\n *\n * Rules (in priority order):\n *  - MEMORY.md anywhere in memory/  → 'evergreen'\n *  - YYYY-MM-DD.md in memory/       → 'daily'\n *  - anything else in memory/       → 'topic'\n *  - anything in Notes/             → 'session'\n */\nexport function detectTier(\n  relativePath: string,\n): \"evergreen\" | \"daily\" | \"topic\" | \"session\" {\n  // Normalise to forward slashes and strip leading ./\n  const p = relativePath.replace(/\\\\/g, \"/\").replace(/^\\.\\//, \"\");\n\n  // Notes directory → session tier\n  if (p.startsWith(\"Notes/\") || p === \"Notes\") {\n    return \"session\";\n  }\n\n  const fileName = basename(p);\n\n  // MEMORY.md (case-sensitive match) → evergreen\n  if (fileName === \"MEMORY.md\") {\n    return \"evergreen\";\n  }\n\n  // YYYY-MM-DD.md → daily\n  if (/^\\d{4}-\\d{2}-\\d{2}\\.md$/.test(fileName)) {\n    return \"daily\";\n  }\n\n  // Default for memory/ files\n  return \"topic\";\n}\n\n// ---------------------------------------------------------------------------\n// Chunk ID generation\n// ---------------------------------------------------------------------------\n\n/**\n * Generate a deterministic chunk ID from its coordinates.\n * Format: sha256(\"projectId:path:chunkIndex:startLine:endLine\")\n *\n * The chunkIndex (0-based position within the file) is included so that\n * chunks with approximated line numbers (e.g. from splitBySentences) never\n * produce colliding IDs even when multiple chunks share the same startLine/endLine.\n */\nfunction chunkId(\n  projectId: number,\n  path: string,\n  chunkIndex: number,\n  startLine: number,\n  endLine: number,\n): string {\n  return createHash(\"sha256\")\n    .update(`${projectId}:${path}:${chunkIndex}:${startLine}:${endLine}`)\n    .digest(\"hex\");\n}\n\n// ---------------------------------------------------------------------------\n// File hash\n// ---------------------------------------------------------------------------\n\nfunction sha256File(content: string): string {\n  return createHash(\"sha256\").update(content).digest(\"hex\");\n}\n\n// ---------------------------------------------------------------------------\n// Core indexing operations\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file into the federation database.\n *\n * @returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport function indexFile(\n  db: Database,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): boolean {\n  const absPath = join(rootPath, relativePath);\n\n  // Read file content\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    // File unreadable or missing — skip silently\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Check if the file has changed since last index\n  const existing = db\n    .prepare(\n      \"SELECT hash FROM memory_files WHERE project_id = ? AND path = ?\",\n    )\n    .get(projectId, relativePath) as { hash: string } | undefined;\n\n  if (existing?.hash === hash) {\n    // Unchanged — skip\n    return false;\n  }\n\n  // Delete old chunks for this file from both tables\n  const oldChunkIds = db\n    .prepare(\n      \"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\",\n    )\n    .all(projectId, relativePath) as Array<{ id: string }>;\n\n  const deleteFts = db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n  const deleteChunk = db.prepare(\n    \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\",\n  );\n\n  db.transaction(() => {\n    for (const row of oldChunkIds) {\n      deleteFts.run(row.id);\n    }\n    deleteChunk.run(projectId, relativePath);\n  })();\n\n  // Chunk the new content\n  const chunks = chunkMarkdown(content);\n\n  // Insert new chunks into memory_chunks and memory_fts\n  const insertChunk = db.prepare(`\n    INSERT INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n  `);\n\n  const insertFts = db.prepare(`\n    INSERT INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n    VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n  `);\n\n  const upsertFile = db.prepare(`\n    INSERT INTO memory_files (project_id, path, source, tier, hash, mtime, size)\n    VALUES (?, ?, ?, ?, ?, ?, ?)\n    ON CONFLICT(project_id, path) DO UPDATE SET\n      source = excluded.source,\n      tier   = excluded.tier,\n      hash   = excluded.hash,\n      mtime  = excluded.mtime,\n      size   = excluded.size\n  `);\n\n  const updatedAt = Date.now();\n\n  db.transaction(() => {\n    for (let i = 0; i < chunks.length; i++) {\n      const chunk = chunks[i]!;\n      const id = chunkId(projectId, relativePath, i, chunk.startLine, chunk.endLine);\n      insertChunk.run(\n        id,\n        projectId,\n        source,\n        tier,\n        relativePath,\n        chunk.startLine,\n        chunk.endLine,\n        chunk.hash,\n        chunk.text,\n        updatedAt,\n      );\n      insertFts.run(\n        chunk.text,\n        id,\n        projectId,\n        relativePath,\n        source,\n        tier,\n        chunk.startLine,\n        chunk.endLine,\n      );\n    }\n    upsertFile.run(projectId, relativePath, source, tier, hash, mtime, size);\n  })();\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Directory walker\n// ---------------------------------------------------------------------------\n\n/**\n * Safety cap: maximum number of .md files collected per project scan.\n * Prevents runaway scans on huge root paths (e.g. home directory).\n * Projects with more files than this are scanned up to the cap only.\n */\nconst MAX_FILES_PER_PROJECT = 5_000;\n\n/**\n * Maximum recursion depth for directory walks.\n * Prevents deep traversal of large directory trees (e.g. development repos).\n * Depth 0 = the given directory itself (no recursion).\n * Value 6 allows: root → subdirs → sub-subdirs → ... up to 6 levels.\n * Sufficient for memory/, Notes/, and typical docs structures.\n */\nconst MAX_WALK_DEPTH = 6;\n\n/**\n * Recursively collect all .md files under a directory.\n * Returns absolute paths. Stops early if the accumulated count hits the cap\n * or if the recursion depth exceeds MAX_WALK_DEPTH.\n *\n * @param dir    Directory to scan.\n * @param acc    Shared accumulator array (mutated in place for early exit).\n * @param cap    Maximum number of files to collect (across all recursive calls).\n * @param depth  Current recursion depth (0 = the initial call).\n */\nfunction walkMdFiles(\n  dir: string,\n  acc?: string[],\n  cap = MAX_FILES_PER_PROJECT,\n  depth = 0,\n): string[] {\n  const results = acc ?? [];\n  if (!existsSync(dir)) return results;\n  if (results.length >= cap) return results;\n  if (depth > MAX_WALK_DEPTH) return results;\n\n  try {\n    for (const entry of readdirSync(dir, { withFileTypes: true })) {\n      if (results.length >= cap) break;\n      if (entry.isSymbolicLink()) continue;\n      // Skip known junk directories at every recursion depth\n      if (ALWAYS_SKIP_DIRS.has(entry.name)) continue;\n      const full = join(dir, entry.name);\n      if (entry.isDirectory()) {\n        walkMdFiles(full, results, cap, depth + 1);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        results.push(full);\n      }\n    }\n  } catch {\n    // Unreadable directory — skip\n  }\n  return results;\n}\n\n/**\n * Directories to ALWAYS skip, at any depth, during any directory walk.\n * These are build artifacts, dependency trees, and VCS internals that\n * should never be indexed regardless of where they appear in the tree.\n */\nconst ALWAYS_SKIP_DIRS = new Set([\n  // Version control\n  \".git\",\n  // Dependency directories (any language)\n  \"node_modules\",\n  \"vendor\",\n  \"Pods\",              // CocoaPods (iOS/macOS)\n  // Build / compile output\n  \"dist\",\n  \"build\",\n  \"out\",\n  \"DerivedData\",       // Xcode\n  \".next\",             // Next.js\n  // Python virtual environments and caches\n  \".venv\",\n  \"venv\",\n  \"__pycache__\",\n  // General caches\n  \".cache\",\n  \".bun\",\n]);\n\n/**\n * Directories to skip when doing a root-level content scan.\n * These are either already handled by dedicated scans or should never be indexed.\n */\nconst ROOT_SCAN_SKIP_DIRS = new Set([\n  \"memory\",\n  \"Notes\",\n  \".claude\",\n  \".DS_Store\",\n  // Everything in ALWAYS_SKIP_DIRS is also excluded at root level\n  ...ALWAYS_SKIP_DIRS,\n]);\n\n/**\n * Additional directories to skip at the content-scan level (first level below root).\n * These are common macOS/Linux home-directory or repo noise directories that are\n * never meaningful as project content.\n */\nconst CONTENT_SCAN_SKIP_DIRS = new Set([\n  // macOS home directory standard folders\n  \"Library\",\n  \"Applications\",\n  \"Music\",\n  \"Movies\",\n  \"Pictures\",\n  \"Desktop\",\n  \"Downloads\",\n  \"Public\",\n  // Common dev noise\n  \"coverage\",\n  // Everything in ALWAYS_SKIP_DIRS is also excluded at this level\n  ...ALWAYS_SKIP_DIRS,\n]);\n\n/**\n * Recursively collect all .md files under rootPath, excluding directories\n * that are already covered by dedicated scans (memory/, Notes/) and\n * common noise directories (.git, node_modules, etc.).\n *\n * Returns absolute paths for files NOT already handled by the specific scanners.\n * Stops collecting once MAX_FILES_PER_PROJECT is reached.\n */\nfunction walkContentFiles(rootPath: string): string[] {\n  if (!existsSync(rootPath)) return [];\n\n  const results: string[] = [];\n  try {\n    for (const entry of readdirSync(rootPath, { withFileTypes: true })) {\n      if (results.length >= MAX_FILES_PER_PROJECT) break;\n      if (entry.isSymbolicLink()) continue;\n      if (ROOT_SCAN_SKIP_DIRS.has(entry.name)) continue;\n      if (CONTENT_SCAN_SKIP_DIRS.has(entry.name)) continue;\n\n      const full = join(rootPath, entry.name);\n      if (entry.isDirectory()) {\n        walkMdFiles(full, results, MAX_FILES_PER_PROJECT);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        // Skip root-level MEMORY.md — handled by the dedicated evergreen scan\n        if (entry.name !== \"MEMORY.md\") {\n          results.push(full);\n        }\n      }\n    }\n  } catch {\n    // Unreadable directory — skip\n  }\n  return results;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing\n// ---------------------------------------------------------------------------\n\n/**\n * Index all memory, Notes, and content files for a single registered project.\n *\n * Scans:\n *  - {rootPath}/MEMORY.md    → source='memory', tier='evergreen'\n *  - {rootPath}/memory/      → source='memory', tier from detectTier()\n *  - {rootPath}/Notes/       → source='notes',  tier='session'\n *  - {rootPath}/**\\/\\*.md    → source='content', tier='topic'  (all other .md files, recursive)\n *  - {claudeNotesDir}/       → source='notes',  tier='session'  (if set and different)\n *\n * The content scan covers projects like job-discussions where markdown files\n * live in date/topic subdirectories rather than a memory/ folder.  The\n * memory/, Notes/, .git/, and node_modules/ directories are excluded from\n * the content scan to avoid double-indexing.\n *\n * The claudeNotesDir parameter points to ~/.claude/projects/{encoded}/Notes/\n * where Claude Code writes session notes for a given working directory.\n * It is stored on the project row as claude_notes_dir after a registry scan.\n */\n/**\n * Number of files to process before yielding to the event loop inside\n * indexProject. Keeps IPC responsive even while indexing large projects.\n * Lower = more responsive but more overhead. 10 is a good balance.\n */\nconst INDEX_YIELD_EVERY = 10;\n\n/**\n * Returns true if rootPath should skip the recursive content scan.\n *\n * Skips content scanning for:\n *  - The home directory itself or any ancestor (too broad — millions of files)\n *  - Git repositories (code repos — index memory/ and Notes/ only, not all .md files)\n *\n * The content scan is still useful for Obsidian vaults, Notes folders, and\n * other doc-centric project trees where ALL markdown files are meaningful.\n *\n * The memory/, Notes/, and claude_notes_dir scans always run regardless.\n */\nfunction isPathTooBroadForContentScan(rootPath: string): boolean {\n  const normalized = normalize(rootPath);\n  const home = homedir();\n\n  // Skip the home directory itself or any ancestor of home\n  if (home.startsWith(normalized) || normalized === \"/\") {\n    return true;\n  }\n\n  // Skip home directory itself (depth 0)\n  if (normalized.startsWith(home)) {\n    const rel = normalized.slice(home.length).replace(/^\\//, \"\");\n    const depth = rel ? rel.split(\"/\").length : 0;\n    if (depth === 0) return true;\n  }\n\n  // Skip git repositories — content scan is only for doc-centric projects\n  // (Obsidian vaults, knowledge bases). Code repos use memory/ and Notes/ only.\n  if (existsSync(join(normalized, \".git\"))) {\n    return true;\n  }\n\n  return false;\n}\n\nexport async function indexProject(\n  db: Database,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = {\n    filesProcessed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n  };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  // Root-level MEMORY.md\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  // memory/ directory\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  // {rootPath}/Notes/ directory\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 with the standard filename format:\n  // \"NNNN - YYYY-MM-DD - Descriptive Title.md\"\n  // These are small, high-signal chunks that make session titles searchable via BM25 and embeddings.\n  {\n    const SESSION_TITLE_RE = /^(\\d{4})\\s*-\\s*(\\d{4}-\\d{2}-\\d{2})\\s*-\\s*(.+)\\.md$/;\n    const titleInsertChunk = db.prepare(`\n      INSERT OR IGNORE INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n      VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n    const titleInsertFts = db.prepare(`\n      INSERT OR IGNORE INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n      VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const m = SESSION_TITLE_RE.exec(fileName);\n      if (!m) continue;\n      const [, num, date, title] = m;\n      const text = `Session #${num} ${date}: ${title}`;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      db.transaction(() => {\n        titleInsertChunk.run(id, projectId, \"notes\", \"session\", syntheticPath, 0, 0, hash, text, updatedAt);\n        titleInsertFts.run(text, id, projectId, syntheticPath, \"notes\", \"session\", 0, 0);\n      })();\n    }\n  }\n\n  // {rootPath}/**/*.md — all other markdown content (e.g. year/month/topic dirs)\n  // Uses walkContentFiles which skips memory/, Notes/, .git/, node_modules/ etc.\n  // Skip the content scan for paths that are too broad (home dir, filesystem root, etc.)\n  // to avoid runaway directory traversal. Memory and Notes scans above are always safe\n  // because they target specific named subdirectories.\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  // Claude Code session notes directory (~/.claude/projects/{encoded}/Notes/)\n  // Only scan if it is set, exists, and is not the same path as rootPath/Notes/\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 SESSION_TITLE_RE_CLAUDE = /^(\\d{4})\\s*-\\s*(\\d{4}-\\d{2}-\\d{2})\\s*-\\s*(.+)\\.md$/;\n      const updatedAt = Date.now();\n      const titleInsertChunk2 = db.prepare(`\n        INSERT OR IGNORE INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n      `);\n      const titleInsertFts2 = db.prepare(`\n        INSERT OR IGNORE INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n        VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n      `);\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const m = SESSION_TITLE_RE_CLAUDE.exec(fileName);\n        if (!m) continue;\n        const [, num, date, title] = m;\n        const text = `Session #${num} ${date}: ${title}`;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        db.transaction(() => {\n          titleInsertChunk2.run(id, projectId, \"notes\", \"session\", syntheticPath, 0, 0, hash, text, updatedAt);\n          titleInsertFts2.run(text, id, projectId, syntheticPath, \"notes\", \"session\", 0, 0);\n        })();\n      }\n    }\n\n    // Derive the sibling memory/ directory: .../Notes/ → .../memory/\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n\n      // MEMORY.md at the Claude Code project dir level (sibling of Notes/)\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({\n          absPath: claudeMemoryMd,\n          rootBase: claudeProjectDir,\n          source: \"memory\",\n          tier: \"evergreen\",\n        });\n      }\n\n      // memory/ directory sibling of Notes/\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  // Yield after collection phase (which is synchronous) before we start processing\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    // Yield to the event loop periodically so the IPC server stays responsive\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    const changed = indexFile(db, projectId, rootBase, relPath, source, tier);\n\n    if (changed) {\n      // Count chunks created for this file\n      const count = db\n        .prepare(\n          \"SELECT COUNT(*) as n FROM memory_chunks WHERE project_id = ? AND path = ?\",\n        )\n        .get(projectId, relPath) as { n: number };\n\n      result.filesProcessed++;\n      result.chunksCreated += count.n;\n    } else {\n      result.filesSkipped++;\n    }\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing (all registered projects)\n// ---------------------------------------------------------------------------\n\n/**\n * Yield to the Node.js event loop between projects so the IPC server\n * remains responsive during long index runs.\n */\nfunction yieldToEventLoop(): Promise<void> {\n  return new Promise((resolve) => setImmediate(resolve));\n}\n\n/**\n * Index all active projects registered in the registry DB.\n *\n * Async: yields to the event loop between each project so that the daemon's\n * Unix socket server can process IPC requests (e.g. status) while indexing.\n */\nexport async function indexAll(\n  db: Database,\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 = {\n    filesProcessed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n  };\n\n  for (const project of projects) {\n    // Yield before each project so the event loop can drain IPC requests\n    await yieldToEventLoop();\n\n    const r = await indexProject(db, 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\n// ---------------------------------------------------------------------------\n// Embedding generation\n// ---------------------------------------------------------------------------\n\nexport interface EmbedResult {\n  chunksEmbedded: number;\n  chunksSkipped: number;\n}\n\n/**\n * Generate and store embeddings for chunks that do not yet have one.\n *\n * Because better-sqlite3 is synchronous but the embedding pipeline is async,\n * we fetch all unembedded chunk texts first, generate embeddings in batches,\n * and then write them back in a transaction.\n *\n * @param db         Open federation database.\n * @param projectId  Optional — restrict to a specific project.\n * @param batchSize  Number of chunks to embed per round. Default 50.\n * @param onProgress Optional callback called after each batch with running totals.\n */\nexport async function embedChunks(\n  db: Database,\n  projectId?: number,\n  batchSize = 50,\n  onProgress?: (embedded: number, total: number) => void,\n): Promise<EmbedResult> {\n  // Dynamic import — keeps the heavy ML runtime out of the module load path\n  const { generateEmbedding, serializeEmbedding } = await import(\"./embeddings.js\");\n\n  const conditions = [\"embedding IS NULL\"];\n  const params: (string | number)[] = [];\n\n  if (projectId !== undefined) {\n    conditions.push(\"project_id = ?\");\n    params.push(projectId);\n  }\n\n  const where = \"WHERE \" + conditions.join(\" AND \");\n\n  const rows = db\n    .prepare(`SELECT id, text FROM memory_chunks ${where} ORDER BY id`)\n    .all(...params) as Array<{ id: string; text: string }>;\n\n  if (rows.length === 0) {\n    return { chunksEmbedded: 0, chunksSkipped: 0 };\n  }\n\n  const updateStmt = db.prepare(\n    \"UPDATE memory_chunks SET embedding = ? WHERE id = ?\",\n  );\n\n  let embedded = 0;\n  const total = rows.length;\n\n  // Process in batches so progress callbacks are meaningful\n  for (let i = 0; i < rows.length; i += batchSize) {\n    const batch = rows.slice(i, i + batchSize);\n\n    // Generate embeddings for the batch (async — must happen OUTSIDE transaction)\n    const embeddings: Array<{ id: string; blob: Buffer }> = [];\n    for (const row of batch) {\n      const vec = await generateEmbedding(row.text);\n      const blob = serializeEmbedding(vec);\n      embeddings.push({ id: row.id, blob });\n    }\n\n    // Write the batch in a single transaction\n    db.transaction(() => {\n      for (const { id, blob } of embeddings) {\n        updateStmt.run(blob, id);\n      }\n    })();\n\n    embedded += embeddings.length;\n    onProgress?.(embedded, total);\n  }\n\n  return { chunksEmbedded: embedded, chunksSkipped: 0 };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AA2CA,SAAgB,WACd,cAC6C;CAE7C,MAAM,IAAI,aAAa,QAAQ,OAAO,IAAI,CAAC,QAAQ,SAAS,GAAG;AAG/D,KAAI,EAAE,WAAW,SAAS,IAAI,MAAM,QAClC,QAAO;CAGT,MAAM,WAAW,SAAS,EAAE;AAG5B,KAAI,aAAa,YACf,QAAO;AAIT,KAAI,0BAA0B,KAAK,SAAS,CAC1C,QAAO;AAIT,QAAO;;;;;;;;;;AAeT,SAAS,QACP,WACA,MACA,YACA,WACA,SACQ;AACR,QAAO,WAAW,SAAS,CACxB,OAAO,GAAG,UAAU,GAAG,KAAK,GAAG,WAAW,GAAG,UAAU,GAAG,UAAU,CACpE,OAAO,MAAM;;AAOlB,SAAS,WAAW,SAAyB;AAC3C,QAAO,WAAW,SAAS,CAAC,OAAO,QAAQ,CAAC,OAAO,MAAM;;;;;;;AAY3D,SAAgB,UACd,IACA,WACA,UACA,cACA,QACA,MACS;CACT,MAAM,UAAU,KAAK,UAAU,aAAa;CAG5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AAEN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AASlB,KANiB,GACd,QACC,kEACD,CACA,IAAI,WAAW,aAAa,EAEjB,SAAS,KAErB,QAAO;CAIT,MAAM,cAAc,GACjB,QACC,iEACD,CACA,IAAI,WAAW,aAAa;CAE/B,MAAM,YAAY,GAAG,QAAQ,sCAAsC;CACnE,MAAM,cAAc,GAAG,QACrB,8DACD;AAED,IAAG,kBAAkB;AACnB,OAAK,MAAM,OAAO,YAChB,WAAU,IAAI,IAAI,GAAG;AAEvB,cAAY,IAAI,WAAW,aAAa;GACxC,EAAE;CAGJ,MAAM,SAAS,cAAc,QAAQ;CAGrC,MAAM,cAAc,GAAG,QAAQ;;;IAG7B;CAEF,MAAM,YAAY,GAAG,QAAQ;;;IAG3B;CAEF,MAAM,aAAa,GAAG,QAAQ;;;;;;;;;IAS5B;CAEF,MAAM,YAAY,KAAK,KAAK;AAE5B,IAAG,kBAAkB;AACnB,OAAK,IAAI,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;GACtC,MAAM,QAAQ,OAAO;GACrB,MAAM,KAAK,QAAQ,WAAW,cAAc,GAAG,MAAM,WAAW,MAAM,QAAQ;AAC9E,eAAY,IACV,IACA,WACA,QACA,MACA,cACA,MAAM,WACN,MAAM,SACN,MAAM,MACN,MAAM,MACN,UACD;AACD,aAAU,IACR,MAAM,MACN,IACA,WACA,cACA,QACA,MACA,MAAM,WACN,MAAM,QACP;;AAEH,aAAW,IAAI,WAAW,cAAc,QAAQ,MAAM,MAAM,OAAO,KAAK;GACxE,EAAE;AAEJ,QAAO;;;;;;;AAYT,MAAM,wBAAwB;;;;;;;;AAS9B,MAAM,iBAAiB;;;;;;;;;;;AAYvB,SAAS,YACP,KACA,KACA,MAAM,uBACN,QAAQ,GACE;CACV,MAAM,UAAU,OAAO,EAAE;AACzB,KAAI,CAAC,WAAW,IAAI,CAAE,QAAO;AAC7B,KAAI,QAAQ,UAAU,IAAK,QAAO;AAClC,KAAI,QAAQ,eAAgB,QAAO;AAEnC,KAAI;AACF,OAAK,MAAM,SAAS,YAAY,KAAK,EAAE,eAAe,MAAM,CAAC,EAAE;AAC7D,OAAI,QAAQ,UAAU,IAAK;AAC3B,OAAI,MAAM,gBAAgB,CAAE;AAE5B,OAAI,iBAAiB,IAAI,MAAM,KAAK,CAAE;GACtC,MAAM,OAAO,KAAK,KAAK,MAAM,KAAK;AAClC,OAAI,MAAM,aAAa,CACrB,aAAY,MAAM,SAAS,KAAK,QAAQ,EAAE;YACjC,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,CACrD,SAAQ,KAAK,KAAK;;SAGhB;AAGR,QAAO;;;;;;;AAQT,MAAM,mBAAmB,IAAI,IAAI;CAE/B;CAEA;CACA;CACA;CAEA;CACA;CACA;CACA;CACA;CAEA;CACA;CACA;CAEA;CACA;CACD,CAAC;;;;;AAMF,MAAM,sBAAsB,IAAI,IAAI;CAClC;CACA;CACA;CACA;CAEA,GAAG;CACJ,CAAC;;;;;;AAOF,MAAM,yBAAyB,IAAI,IAAI;CAErC;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CAEA;CAEA,GAAG;CACJ,CAAC;;;;;;;;;AAUF,SAAS,iBAAiB,UAA4B;AACpD,KAAI,CAAC,WAAW,SAAS,CAAE,QAAO,EAAE;CAEpC,MAAM,UAAoB,EAAE;AAC5B,KAAI;AACF,OAAK,MAAM,SAAS,YAAY,UAAU,EAAE,eAAe,MAAM,CAAC,EAAE;AAClE,OAAI,QAAQ,UAAU,sBAAuB;AAC7C,OAAI,MAAM,gBAAgB,CAAE;AAC5B,OAAI,oBAAoB,IAAI,MAAM,KAAK,CAAE;AACzC,OAAI,uBAAuB,IAAI,MAAM,KAAK,CAAE;GAE5C,MAAM,OAAO,KAAK,UAAU,MAAM,KAAK;AACvC,OAAI,MAAM,aAAa,CACrB,aAAY,MAAM,SAAS,sBAAsB;YACxC,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,EAErD;QAAI,MAAM,SAAS,YACjB,SAAQ,KAAK,KAAK;;;SAIlB;AAGR,QAAO;;;;;;;;;;;;;;;;;;;;;;;;;;AA+BT,MAAM,oBAAoB;;;;;;;;;;;;;AAc1B,SAAS,6BAA6B,UAA2B;CAC/D,MAAM,aAAa,UAAU,SAAS;CACtC,MAAM,OAAO,SAAS;AAGtB,KAAI,KAAK,WAAW,WAAW,IAAI,eAAe,IAChD,QAAO;AAIT,KAAI,WAAW,WAAW,KAAK,EAAE;EAC/B,MAAM,MAAM,WAAW,MAAM,KAAK,OAAO,CAAC,QAAQ,OAAO,GAAG;AAE5D,OADc,MAAM,IAAI,MAAM,IAAI,CAAC,SAAS,OAC9B,EAAG,QAAO;;AAK1B,KAAI,WAAW,KAAK,YAAY,OAAO,CAAC,CACtC,QAAO;AAGT,QAAO;;AAGT,eAAsB,aACpB,IACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAC1B,gBAAgB;EAChB,eAAe;EACf,cAAc;EACf;CAED,MAAM,eAA2F,EAAE;CAGnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAIvG,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;;CAI5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAMtF;EACE,MAAM,mBAAmB;EACzB,MAAM,mBAAmB,GAAG,QAAQ;;;MAGlC;EACF,MAAM,iBAAiB,GAAG,QAAQ;;;MAGhC;EACF,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAC3C,MAAM,WAAW,SAAS,QAAQ;GAClC,MAAM,IAAI,iBAAiB,KAAK,SAAS;AACzC,OAAI,CAAC,EAAG;GACR,MAAM,GAAG,KAAK,MAAM,SAAS;GAC7B,MAAM,OAAO,YAAY,IAAI,GAAG,KAAK,IAAI;GAEzC,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GACjC,MAAM,KAAK,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;GACrD,MAAM,OAAO,WAAW,KAAK;AAC7B,MAAG,kBAAkB;AACnB,qBAAiB,IAAI,IAAI,WAAW,SAAS,WAAW,eAAe,GAAG,GAAG,MAAM,MAAM,UAAU;AACnG,mBAAe,IAAI,MAAM,IAAI,WAAW,eAAe,SAAS,WAAW,GAAG,EAAE;KAChF,EAAE;;;AASR,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAMxF,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,0BAA0B;GAChC,MAAM,YAAY,KAAK,KAAK;GAC5B,MAAM,oBAAoB,GAAG,QAAQ;;;QAGnC;GACF,MAAM,kBAAkB,GAAG,QAAQ;;;QAGjC;AACF,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IACjD,MAAM,WAAW,SAAS,QAAQ;IAClC,MAAM,IAAI,wBAAwB,KAAK,SAAS;AAChD,QAAI,CAAC,EAAG;IACR,MAAM,GAAG,KAAK,MAAM,SAAS;IAC7B,MAAM,OAAO,YAAY,IAAI,GAAG,KAAK,IAAI;IAEzC,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IACjC,MAAM,KAAK,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IACrD,MAAM,OAAO,WAAW,KAAK;AAC7B,OAAG,kBAAkB;AACnB,uBAAkB,IAAI,IAAI,WAAW,SAAS,WAAW,eAAe,GAAG,GAAG,MAAM,MAAM,UAAU;AACpG,qBAAgB,IAAI,MAAM,IAAI,WAAW,eAAe,SAAS,WAAW,GAAG,EAAE;MACjF,EAAE;;;AAKR,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;GAGxD,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAChB,SAAS;IACT,UAAU;IACV,QAAQ;IACR,MAAM;IACP,CAAC;AAIJ,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;;;;AAMxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAE9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAG3C,MAFgB,UAAU,IAAI,WAAW,UAAU,SAAS,QAAQ,KAAK,EAE5D;GAEX,MAAM,QAAQ,GACX,QACC,4EACD,CACA,IAAI,WAAW,QAAQ;AAE1B,UAAO;AACP,UAAO,iBAAiB,MAAM;QAE9B,QAAO;;AAIX,QAAO;;;;;;AAWT,SAAS,mBAAkC;AACzC,QAAO,IAAI,SAAS,YAAY,aAAa,QAAQ,CAAC;;;;;;;;AASxD,eAAsB,SACpB,IACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAC1B,gBAAgB;EAChB,eAAe;EACf,cAAc;EACf;AAED,MAAK,MAAM,WAAW,UAAU;AAE9B,QAAM,kBAAkB;EAExB,MAAM,IAAI,MAAM,aAAa,IAAI,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzF,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ;;;;;;;;;;;;;;AAwBtD,eAAsB,YACpB,IACA,WACA,YAAY,IACZ,YACsB;CAEtB,MAAM,EAAE,mBAAmB,uBAAuB,MAAM,OAAO;CAE/D,MAAM,aAAa,CAAC,oBAAoB;CACxC,MAAM,SAA8B,EAAE;AAEtC,KAAI,cAAc,QAAW;AAC3B,aAAW,KAAK,iBAAiB;AACjC,SAAO,KAAK,UAAU;;CAGxB,MAAM,QAAQ,WAAW,WAAW,KAAK,QAAQ;CAEjD,MAAM,OAAO,GACV,QAAQ,sCAAsC,MAAM,cAAc,CAClE,IAAI,GAAG,OAAO;AAEjB,KAAI,KAAK,WAAW,EAClB,QAAO;EAAE,gBAAgB;EAAG,eAAe;EAAG;CAGhD,MAAM,aAAa,GAAG,QACpB,sDACD;CAED,IAAI,WAAW;CACf,MAAM,QAAQ,KAAK;AAGnB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,WAAW;EAC/C,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,UAAU;EAG1C,MAAM,aAAkD,EAAE;AAC1D,OAAK,MAAM,OAAO,OAAO;GAEvB,MAAM,OAAO,mBADD,MAAM,kBAAkB,IAAI,KAAK,CACT;AACpC,cAAW,KAAK;IAAE,IAAI,IAAI;IAAI;IAAM,CAAC;;AAIvC,KAAG,kBAAkB;AACnB,QAAK,MAAM,EAAE,IAAI,UAAU,WACzB,YAAW,IAAI,MAAM,GAAG;IAE1B,EAAE;AAEJ,cAAY,WAAW;AACvB,eAAa,UAAU,MAAM;;AAG/B,QAAO;EAAE,gBAAgB;EAAU,eAAe;EAAG"}
+{"version":3,"file":"indexer-CMPOiY1r.mjs","names":[],"sources":["../src/memory/indexer.ts"],"sourcesContent":["/**\n * File indexer for the PAI federation memory engine.\n *\n * Scans project memory/ and Notes/ directories, chunks markdown files, and\n * inserts the resulting chunks into federation.db for BM25 search.\n *\n * Change detection: files whose SHA-256 hash has not changed since the last\n * index run are skipped, keeping incremental re-indexing fast.\n *\n * Phase 2.5: adds embedChunks() for generating vector embeddings on indexed\n * chunks that do not yet have an embedding stored.\n */\n\nimport { createHash } from \"node:crypto\";\nimport { readFileSync, statSync, readdirSync, existsSync } from \"node:fs\";\nimport { join, relative, basename, normalize } from \"node:path\";\nimport { homedir } from \"node:os\";\nimport type { Database } from \"better-sqlite3\";\nimport { chunkMarkdown } from \"./chunker.js\";\n\n// ---------------------------------------------------------------------------\n// Types\n// ---------------------------------------------------------------------------\n\nexport interface IndexResult {\n  filesProcessed: number;\n  chunksCreated: number;\n  filesSkipped: number;\n}\n\n// ---------------------------------------------------------------------------\n// Tier detection\n// ---------------------------------------------------------------------------\n\n/**\n * Classify a relative file path into one of the four memory tiers.\n *\n * Rules (in priority order):\n *  - MEMORY.md anywhere in memory/  → 'evergreen'\n *  - YYYY-MM-DD.md in memory/       → 'daily'\n *  - anything else in memory/       → 'topic'\n *  - anything in Notes/             → 'session'\n */\nexport function detectTier(\n  relativePath: string,\n): \"evergreen\" | \"daily\" | \"topic\" | \"session\" {\n  // Normalise to forward slashes and strip leading ./\n  const p = relativePath.replace(/\\\\/g, \"/\").replace(/^\\.\\//, \"\");\n\n  // Notes directory → session tier\n  if (p.startsWith(\"Notes/\") || p === \"Notes\") {\n    return \"session\";\n  }\n\n  const fileName = basename(p);\n\n  // MEMORY.md (case-sensitive match) → evergreen\n  if (fileName === \"MEMORY.md\") {\n    return \"evergreen\";\n  }\n\n  // YYYY-MM-DD.md → daily\n  if (/^\\d{4}-\\d{2}-\\d{2}\\.md$/.test(fileName)) {\n    return \"daily\";\n  }\n\n  // Default for memory/ files\n  return \"topic\";\n}\n\n// ---------------------------------------------------------------------------\n// Chunk ID generation\n// ---------------------------------------------------------------------------\n\n/**\n * Generate a deterministic chunk ID from its coordinates.\n * Format: sha256(\"projectId:path:chunkIndex:startLine:endLine\")\n *\n * The chunkIndex (0-based position within the file) is included so that\n * chunks with approximated line numbers (e.g. from splitBySentences) never\n * produce colliding IDs even when multiple chunks share the same startLine/endLine.\n */\nfunction chunkId(\n  projectId: number,\n  path: string,\n  chunkIndex: number,\n  startLine: number,\n  endLine: number,\n): string {\n  return createHash(\"sha256\")\n    .update(`${projectId}:${path}:${chunkIndex}:${startLine}:${endLine}`)\n    .digest(\"hex\");\n}\n\n// ---------------------------------------------------------------------------\n// File hash\n// ---------------------------------------------------------------------------\n\nfunction sha256File(content: string): string {\n  return createHash(\"sha256\").update(content).digest(\"hex\");\n}\n\n// ---------------------------------------------------------------------------\n// Core indexing operations\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file into the federation database.\n *\n * @returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport function indexFile(\n  db: Database,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): boolean {\n  const absPath = join(rootPath, relativePath);\n\n  // Read file content\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    // File unreadable or missing — skip silently\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Check if the file has changed since last index\n  const existing = db\n    .prepare(\n      \"SELECT hash FROM memory_files WHERE project_id = ? AND path = ?\",\n    )\n    .get(projectId, relativePath) as { hash: string } | undefined;\n\n  if (existing?.hash === hash) {\n    // Unchanged — skip\n    return false;\n  }\n\n  // Delete old chunks for this file from both tables\n  const oldChunkIds = db\n    .prepare(\n      \"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\",\n    )\n    .all(projectId, relativePath) as Array<{ id: string }>;\n\n  const deleteFts = db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n  const deleteChunk = db.prepare(\n    \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\",\n  );\n\n  db.transaction(() => {\n    for (const row of oldChunkIds) {\n      deleteFts.run(row.id);\n    }\n    deleteChunk.run(projectId, relativePath);\n  })();\n\n  // Chunk the new content\n  const chunks = chunkMarkdown(content);\n\n  // Insert new chunks into memory_chunks and memory_fts\n  const insertChunk = db.prepare(`\n    INSERT INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n  `);\n\n  const insertFts = db.prepare(`\n    INSERT INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n    VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n  `);\n\n  const upsertFile = db.prepare(`\n    INSERT INTO memory_files (project_id, path, source, tier, hash, mtime, size)\n    VALUES (?, ?, ?, ?, ?, ?, ?)\n    ON CONFLICT(project_id, path) DO UPDATE SET\n      source = excluded.source,\n      tier   = excluded.tier,\n      hash   = excluded.hash,\n      mtime  = excluded.mtime,\n      size   = excluded.size\n  `);\n\n  const updatedAt = Date.now();\n\n  db.transaction(() => {\n    for (let i = 0; i < chunks.length; i++) {\n      const chunk = chunks[i]!;\n      const id = chunkId(projectId, relativePath, i, chunk.startLine, chunk.endLine);\n      insertChunk.run(\n        id,\n        projectId,\n        source,\n        tier,\n        relativePath,\n        chunk.startLine,\n        chunk.endLine,\n        chunk.hash,\n        chunk.text,\n        updatedAt,\n      );\n      insertFts.run(\n        chunk.text,\n        id,\n        projectId,\n        relativePath,\n        source,\n        tier,\n        chunk.startLine,\n        chunk.endLine,\n      );\n    }\n    upsertFile.run(projectId, relativePath, source, tier, hash, mtime, size);\n  })();\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Directory walker\n// ---------------------------------------------------------------------------\n\n/**\n * Safety cap: maximum number of .md files collected per project scan.\n * Prevents runaway scans on huge root paths (e.g. home directory).\n * Projects with more files than this are scanned up to the cap only.\n */\nconst MAX_FILES_PER_PROJECT = 5_000;\n\n/**\n * Maximum recursion depth for directory walks.\n * Prevents deep traversal of large directory trees (e.g. development repos).\n * Depth 0 = the given directory itself (no recursion).\n * Value 6 allows: root → subdirs → sub-subdirs → ... up to 6 levels.\n * Sufficient for memory/, Notes/, and typical docs structures.\n */\nconst MAX_WALK_DEPTH = 6;\n\n/**\n * Recursively collect all .md files under a directory.\n * Returns absolute paths. Stops early if the accumulated count hits the cap\n * or if the recursion depth exceeds MAX_WALK_DEPTH.\n *\n * @param dir    Directory to scan.\n * @param acc    Shared accumulator array (mutated in place for early exit).\n * @param cap    Maximum number of files to collect (across all recursive calls).\n * @param depth  Current recursion depth (0 = the initial call).\n */\nfunction walkMdFiles(\n  dir: string,\n  acc?: string[],\n  cap = MAX_FILES_PER_PROJECT,\n  depth = 0,\n): string[] {\n  const results = acc ?? [];\n  if (!existsSync(dir)) return results;\n  if (results.length >= cap) return results;\n  if (depth > MAX_WALK_DEPTH) return results;\n\n  try {\n    for (const entry of readdirSync(dir, { withFileTypes: true })) {\n      if (results.length >= cap) break;\n      if (entry.isSymbolicLink()) continue;\n      // Skip known junk directories at every recursion depth\n      if (ALWAYS_SKIP_DIRS.has(entry.name)) continue;\n      const full = join(dir, entry.name);\n      if (entry.isDirectory()) {\n        walkMdFiles(full, results, cap, depth + 1);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        results.push(full);\n      }\n    }\n  } catch {\n    // Unreadable directory — skip\n  }\n  return results;\n}\n\n/**\n * Directories to ALWAYS skip, at any depth, during any directory walk.\n * These are build artifacts, dependency trees, and VCS internals that\n * should never be indexed regardless of where they appear in the tree.\n */\nconst ALWAYS_SKIP_DIRS = new Set([\n  // Version control\n  \".git\",\n  // Dependency directories (any language)\n  \"node_modules\",\n  \"vendor\",\n  \"Pods\",              // CocoaPods (iOS/macOS)\n  // Build / compile output\n  \"dist\",\n  \"build\",\n  \"out\",\n  \"DerivedData\",       // Xcode\n  \".next\",             // Next.js\n  // Python virtual environments and caches\n  \".venv\",\n  \"venv\",\n  \"__pycache__\",\n  // General caches\n  \".cache\",\n  \".bun\",\n]);\n\n/**\n * Directories to skip when doing a root-level content scan.\n * These are either already handled by dedicated scans or should never be indexed.\n */\nconst ROOT_SCAN_SKIP_DIRS = new Set([\n  \"memory\",\n  \"Notes\",\n  \".claude\",\n  \".DS_Store\",\n  // Everything in ALWAYS_SKIP_DIRS is also excluded at root level\n  ...ALWAYS_SKIP_DIRS,\n]);\n\n/**\n * Additional directories to skip at the content-scan level (first level below root).\n * These are common macOS/Linux home-directory or repo noise directories that are\n * never meaningful as project content.\n */\nconst CONTENT_SCAN_SKIP_DIRS = new Set([\n  // macOS home directory standard folders\n  \"Library\",\n  \"Applications\",\n  \"Music\",\n  \"Movies\",\n  \"Pictures\",\n  \"Desktop\",\n  \"Downloads\",\n  \"Public\",\n  // Common dev noise\n  \"coverage\",\n  // Everything in ALWAYS_SKIP_DIRS is also excluded at this level\n  ...ALWAYS_SKIP_DIRS,\n]);\n\n/**\n * Recursively collect all .md files under rootPath, excluding directories\n * that are already covered by dedicated scans (memory/, Notes/) and\n * common noise directories (.git, node_modules, etc.).\n *\n * Returns absolute paths for files NOT already handled by the specific scanners.\n * Stops collecting once MAX_FILES_PER_PROJECT is reached.\n */\nfunction walkContentFiles(rootPath: string): string[] {\n  if (!existsSync(rootPath)) return [];\n\n  const results: string[] = [];\n  try {\n    for (const entry of readdirSync(rootPath, { withFileTypes: true })) {\n      if (results.length >= MAX_FILES_PER_PROJECT) break;\n      if (entry.isSymbolicLink()) continue;\n      if (ROOT_SCAN_SKIP_DIRS.has(entry.name)) continue;\n      if (CONTENT_SCAN_SKIP_DIRS.has(entry.name)) continue;\n\n      const full = join(rootPath, entry.name);\n      if (entry.isDirectory()) {\n        walkMdFiles(full, results, MAX_FILES_PER_PROJECT);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        // Skip root-level MEMORY.md — handled by the dedicated evergreen scan\n        if (entry.name !== \"MEMORY.md\") {\n          results.push(full);\n        }\n      }\n    }\n  } catch {\n    // Unreadable directory — skip\n  }\n  return results;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing\n// ---------------------------------------------------------------------------\n\n/**\n * Index all memory, Notes, and content files for a single registered project.\n *\n * Scans:\n *  - {rootPath}/MEMORY.md    → source='memory', tier='evergreen'\n *  - {rootPath}/memory/      → source='memory', tier from detectTier()\n *  - {rootPath}/Notes/       → source='notes',  tier='session'\n *  - {rootPath}/**\\/\\*.md    → source='content', tier='topic'  (all other .md files, recursive)\n *  - {claudeNotesDir}/       → source='notes',  tier='session'  (if set and different)\n *\n * The content scan covers projects like job-discussions where markdown files\n * live in date/topic subdirectories rather than a memory/ folder.  The\n * memory/, Notes/, .git/, and node_modules/ directories are excluded from\n * the content scan to avoid double-indexing.\n *\n * The claudeNotesDir parameter points to ~/.claude/projects/{encoded}/Notes/\n * where Claude Code writes session notes for a given working directory.\n * It is stored on the project row as claude_notes_dir after a registry scan.\n */\n/**\n * Number of files to process before yielding to the event loop inside\n * indexProject. Keeps IPC responsive even while indexing large projects.\n * Lower = more responsive but more overhead. 10 is a good balance.\n */\nconst INDEX_YIELD_EVERY = 10;\n\n/**\n * Returns true if rootPath should skip the recursive content scan.\n *\n * Skips content scanning for:\n *  - The home directory itself or any ancestor (too broad — millions of files)\n *  - Git repositories (code repos — index memory/ and Notes/ only, not all .md files)\n *\n * The content scan is still useful for Obsidian vaults, Notes folders, and\n * other doc-centric project trees where ALL markdown files are meaningful.\n *\n * The memory/, Notes/, and claude_notes_dir scans always run regardless.\n */\nfunction isPathTooBroadForContentScan(rootPath: string): boolean {\n  const normalized = normalize(rootPath);\n  const home = homedir();\n\n  // Skip the home directory itself or any ancestor of home\n  if (home.startsWith(normalized) || normalized === \"/\") {\n    return true;\n  }\n\n  // Skip home directory itself (depth 0)\n  if (normalized.startsWith(home)) {\n    const rel = normalized.slice(home.length).replace(/^\\//, \"\");\n    const depth = rel ? rel.split(\"/\").length : 0;\n    if (depth === 0) return true;\n  }\n\n  // Skip git repositories — content scan is only for doc-centric projects\n  // (Obsidian vaults, knowledge bases). Code repos use memory/ and Notes/ only.\n  if (existsSync(join(normalized, \".git\"))) {\n    return true;\n  }\n\n  return false;\n}\n\nexport async function indexProject(\n  db: Database,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = {\n    filesProcessed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n  };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  // Root-level MEMORY.md\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  // memory/ directory\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  // {rootPath}/Notes/ directory\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 with the standard filename format:\n  // \"NNNN - YYYY-MM-DD - Descriptive Title.md\"\n  // These are small, high-signal chunks that make session titles searchable via BM25 and embeddings.\n  {\n    const SESSION_TITLE_RE = /^(\\d{4})\\s*-\\s*(\\d{4}-\\d{2}-\\d{2})\\s*-\\s*(.+)\\.md$/;\n    const titleInsertChunk = db.prepare(`\n      INSERT OR IGNORE INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n      VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n    const titleInsertFts = db.prepare(`\n      INSERT OR IGNORE INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n      VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n    `);\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const m = SESSION_TITLE_RE.exec(fileName);\n      if (!m) continue;\n      const [, num, date, title] = m;\n      const text = `Session #${num} ${date}: ${title}`;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      db.transaction(() => {\n        titleInsertChunk.run(id, projectId, \"notes\", \"session\", syntheticPath, 0, 0, hash, text, updatedAt);\n        titleInsertFts.run(text, id, projectId, syntheticPath, \"notes\", \"session\", 0, 0);\n      })();\n    }\n  }\n\n  // {rootPath}/**/*.md — all other markdown content (e.g. year/month/topic dirs)\n  // Uses walkContentFiles which skips memory/, Notes/, .git/, node_modules/ etc.\n  // Skip the content scan for paths that are too broad (home dir, filesystem root, etc.)\n  // to avoid runaway directory traversal. Memory and Notes scans above are always safe\n  // because they target specific named subdirectories.\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  // Claude Code session notes directory (~/.claude/projects/{encoded}/Notes/)\n  // Only scan if it is set, exists, and is not the same path as rootPath/Notes/\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 SESSION_TITLE_RE_CLAUDE = /^(\\d{4})\\s*-\\s*(\\d{4}-\\d{2}-\\d{2})\\s*-\\s*(.+)\\.md$/;\n      const updatedAt = Date.now();\n      const titleInsertChunk2 = db.prepare(`\n        INSERT OR IGNORE INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)\n      `);\n      const titleInsertFts2 = db.prepare(`\n        INSERT OR IGNORE INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n        VALUES (?, ?, ?, ?, ?, ?, ?, ?)\n      `);\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const m = SESSION_TITLE_RE_CLAUDE.exec(fileName);\n        if (!m) continue;\n        const [, num, date, title] = m;\n        const text = `Session #${num} ${date}: ${title}`;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        db.transaction(() => {\n          titleInsertChunk2.run(id, projectId, \"notes\", \"session\", syntheticPath, 0, 0, hash, text, updatedAt);\n          titleInsertFts2.run(text, id, projectId, syntheticPath, \"notes\", \"session\", 0, 0);\n        })();\n      }\n    }\n\n    // Derive the sibling memory/ directory: .../Notes/ → .../memory/\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n\n      // MEMORY.md at the Claude Code project dir level (sibling of Notes/)\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({\n          absPath: claudeMemoryMd,\n          rootBase: claudeProjectDir,\n          source: \"memory\",\n          tier: \"evergreen\",\n        });\n      }\n\n      // memory/ directory sibling of Notes/\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  // Yield after collection phase (which is synchronous) before we start processing\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    // Yield to the event loop periodically so the IPC server stays responsive\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    const changed = indexFile(db, projectId, rootBase, relPath, source, tier);\n\n    if (changed) {\n      // Count chunks created for this file\n      const count = db\n        .prepare(\n          \"SELECT COUNT(*) as n FROM memory_chunks WHERE project_id = ? AND path = ?\",\n        )\n        .get(projectId, relPath) as { n: number };\n\n      result.filesProcessed++;\n      result.chunksCreated += count.n;\n    } else {\n      result.filesSkipped++;\n    }\n  }\n\n  // ---------------------------------------------------------------------------\n  // Prune stale paths: remove DB entries for files that no longer exist on disk.\n  // This handles renames, moves, and deletions — the indexer only adds/updates,\n  // so without pruning, old paths accumulate forever.\n  // ---------------------------------------------------------------------------\n\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  // Query all distinct paths in memory_chunks for this project\n  const dbChunkPaths = db\n    .prepare(\"SELECT DISTINCT path FROM memory_chunks WHERE project_id = ?\")\n    .all(projectId) as Array<{ path: string }>;\n\n  const stalePaths: string[] = [];\n  for (const row of dbChunkPaths) {\n    // Synthetic title paths (ending in \"::title\") are live if their base file is live\n    const basePath = row.path.endsWith(\"::title\")\n      ? row.path.slice(0, -\"::title\".length)\n      : row.path;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(row.path);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    const deleteChunksFts = db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n    const deleteChunks = db.prepare(\n      \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\",\n    );\n    const deleteFile = db.prepare(\n      \"DELETE FROM memory_files WHERE project_id = ? AND path = ?\",\n    );\n\n    db.transaction(() => {\n      for (const stalePath of stalePaths) {\n        const chunkIds = db\n          .prepare(\"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\")\n          .all(projectId, stalePath) as Array<{ id: string }>;\n        for (const { id } of chunkIds) {\n          deleteChunksFts.run(id);\n        }\n        deleteChunks.run(projectId, stalePath);\n        deleteFile.run(projectId, stalePath);\n      }\n    })();\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing (all registered projects)\n// ---------------------------------------------------------------------------\n\n/**\n * Yield to the Node.js event loop between projects so the IPC server\n * remains responsive during long index runs.\n */\nfunction yieldToEventLoop(): Promise<void> {\n  return new Promise((resolve) => setImmediate(resolve));\n}\n\n/**\n * Index all active projects registered in the registry DB.\n *\n * Async: yields to the event loop between each project so that the daemon's\n * Unix socket server can process IPC requests (e.g. status) while indexing.\n */\nexport async function indexAll(\n  db: Database,\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 = {\n    filesProcessed: 0,\n    chunksCreated: 0,\n    filesSkipped: 0,\n  };\n\n  for (const project of projects) {\n    // Yield before each project so the event loop can drain IPC requests\n    await yieldToEventLoop();\n\n    const r = await indexProject(db, 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\n// ---------------------------------------------------------------------------\n// Embedding generation\n// ---------------------------------------------------------------------------\n\nexport interface EmbedResult {\n  chunksEmbedded: number;\n  chunksSkipped: number;\n}\n\n/**\n * Generate and store embeddings for chunks that do not yet have one.\n *\n * Because better-sqlite3 is synchronous but the embedding pipeline is async,\n * we fetch all unembedded chunk texts first, generate embeddings in batches,\n * and then write them back in a transaction.\n *\n * @param db         Open federation database.\n * @param projectId  Optional — restrict to a specific project.\n * @param batchSize  Number of chunks to embed per round. Default 50.\n * @param onProgress Optional callback called after each batch with running totals.\n */\nexport async function embedChunks(\n  db: Database,\n  projectId?: number,\n  batchSize = 50,\n  onProgress?: (embedded: number, total: number) => void,\n): Promise<EmbedResult> {\n  // Dynamic import — keeps the heavy ML runtime out of the module load path\n  const { generateEmbedding, serializeEmbedding } = await import(\"./embeddings.js\");\n\n  const conditions = [\"embedding IS NULL\"];\n  const params: (string | number)[] = [];\n\n  if (projectId !== undefined) {\n    conditions.push(\"project_id = ?\");\n    params.push(projectId);\n  }\n\n  const where = \"WHERE \" + conditions.join(\" AND \");\n\n  const rows = db\n    .prepare(`SELECT id, text FROM memory_chunks ${where} ORDER BY id`)\n    .all(...params) as Array<{ id: string; text: string }>;\n\n  if (rows.length === 0) {\n    return { chunksEmbedded: 0, chunksSkipped: 0 };\n  }\n\n  const updateStmt = db.prepare(\n    \"UPDATE memory_chunks SET embedding = ? WHERE id = ?\",\n  );\n\n  let embedded = 0;\n  const total = rows.length;\n\n  // Process in batches so progress callbacks are meaningful\n  for (let i = 0; i < rows.length; i += batchSize) {\n    const batch = rows.slice(i, i + batchSize);\n\n    // Generate embeddings for the batch (async — must happen OUTSIDE transaction)\n    const embeddings: Array<{ id: string; blob: Buffer }> = [];\n    for (const row of batch) {\n      const vec = await generateEmbedding(row.text);\n      const blob = serializeEmbedding(vec);\n      embeddings.push({ id: row.id, blob });\n    }\n\n    // Write the batch in a single transaction\n    db.transaction(() => {\n      for (const { id, blob } of embeddings) {\n        updateStmt.run(blob, id);\n      }\n    })();\n\n    embedded += embeddings.length;\n    onProgress?.(embedded, total);\n  }\n\n  return { chunksEmbedded: embedded, chunksSkipped: 0 };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AA2CA,SAAgB,WACd,cAC6C;CAE7C,MAAM,IAAI,aAAa,QAAQ,OAAO,IAAI,CAAC,QAAQ,SAAS,GAAG;AAG/D,KAAI,EAAE,WAAW,SAAS,IAAI,MAAM,QAClC,QAAO;CAGT,MAAM,WAAW,SAAS,EAAE;AAG5B,KAAI,aAAa,YACf,QAAO;AAIT,KAAI,0BAA0B,KAAK,SAAS,CAC1C,QAAO;AAIT,QAAO;;;;;;;;;;AAeT,SAAS,QACP,WACA,MACA,YACA,WACA,SACQ;AACR,QAAO,WAAW,SAAS,CACxB,OAAO,GAAG,UAAU,GAAG,KAAK,GAAG,WAAW,GAAG,UAAU,GAAG,UAAU,CACpE,OAAO,MAAM;;AAOlB,SAAS,WAAW,SAAyB;AAC3C,QAAO,WAAW,SAAS,CAAC,OAAO,QAAQ,CAAC,OAAO,MAAM;;;;;;;AAY3D,SAAgB,UACd,IACA,WACA,UACA,cACA,QACA,MACS;CACT,MAAM,UAAU,KAAK,UAAU,aAAa;CAG5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AAEN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AASlB,KANiB,GACd,QACC,kEACD,CACA,IAAI,WAAW,aAAa,EAEjB,SAAS,KAErB,QAAO;CAIT,MAAM,cAAc,GACjB,QACC,iEACD,CACA,IAAI,WAAW,aAAa;CAE/B,MAAM,YAAY,GAAG,QAAQ,sCAAsC;CACnE,MAAM,cAAc,GAAG,QACrB,8DACD;AAED,IAAG,kBAAkB;AACnB,OAAK,MAAM,OAAO,YAChB,WAAU,IAAI,IAAI,GAAG;AAEvB,cAAY,IAAI,WAAW,aAAa;GACxC,EAAE;CAGJ,MAAM,SAAS,cAAc,QAAQ;CAGrC,MAAM,cAAc,GAAG,QAAQ;;;IAG7B;CAEF,MAAM,YAAY,GAAG,QAAQ;;;IAG3B;CAEF,MAAM,aAAa,GAAG,QAAQ;;;;;;;;;IAS5B;CAEF,MAAM,YAAY,KAAK,KAAK;AAE5B,IAAG,kBAAkB;AACnB,OAAK,IAAI,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;GACtC,MAAM,QAAQ,OAAO;GACrB,MAAM,KAAK,QAAQ,WAAW,cAAc,GAAG,MAAM,WAAW,MAAM,QAAQ;AAC9E,eAAY,IACV,IACA,WACA,QACA,MACA,cACA,MAAM,WACN,MAAM,SACN,MAAM,MACN,MAAM,MACN,UACD;AACD,aAAU,IACR,MAAM,MACN,IACA,WACA,cACA,QACA,MACA,MAAM,WACN,MAAM,QACP;;AAEH,aAAW,IAAI,WAAW,cAAc,QAAQ,MAAM,MAAM,OAAO,KAAK;GACxE,EAAE;AAEJ,QAAO;;;;;;;AAYT,MAAM,wBAAwB;;;;;;;;AAS9B,MAAM,iBAAiB;;;;;;;;;;;AAYvB,SAAS,YACP,KACA,KACA,MAAM,uBACN,QAAQ,GACE;CACV,MAAM,UAAU,OAAO,EAAE;AACzB,KAAI,CAAC,WAAW,IAAI,CAAE,QAAO;AAC7B,KAAI,QAAQ,UAAU,IAAK,QAAO;AAClC,KAAI,QAAQ,eAAgB,QAAO;AAEnC,KAAI;AACF,OAAK,MAAM,SAAS,YAAY,KAAK,EAAE,eAAe,MAAM,CAAC,EAAE;AAC7D,OAAI,QAAQ,UAAU,IAAK;AAC3B,OAAI,MAAM,gBAAgB,CAAE;AAE5B,OAAI,iBAAiB,IAAI,MAAM,KAAK,CAAE;GACtC,MAAM,OAAO,KAAK,KAAK,MAAM,KAAK;AAClC,OAAI,MAAM,aAAa,CACrB,aAAY,MAAM,SAAS,KAAK,QAAQ,EAAE;YACjC,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,CACrD,SAAQ,KAAK,KAAK;;SAGhB;AAGR,QAAO;;;;;;;AAQT,MAAM,mBAAmB,IAAI,IAAI;CAE/B;CAEA;CACA;CACA;CAEA;CACA;CACA;CACA;CACA;CAEA;CACA;CACA;CAEA;CACA;CACD,CAAC;;;;;AAMF,MAAM,sBAAsB,IAAI,IAAI;CAClC;CACA;CACA;CACA;CAEA,GAAG;CACJ,CAAC;;;;;;AAOF,MAAM,yBAAyB,IAAI,IAAI;CAErC;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CAEA;CAEA,GAAG;CACJ,CAAC;;;;;;;;;AAUF,SAAS,iBAAiB,UAA4B;AACpD,KAAI,CAAC,WAAW,SAAS,CAAE,QAAO,EAAE;CAEpC,MAAM,UAAoB,EAAE;AAC5B,KAAI;AACF,OAAK,MAAM,SAAS,YAAY,UAAU,EAAE,eAAe,MAAM,CAAC,EAAE;AAClE,OAAI,QAAQ,UAAU,sBAAuB;AAC7C,OAAI,MAAM,gBAAgB,CAAE;AAC5B,OAAI,oBAAoB,IAAI,MAAM,KAAK,CAAE;AACzC,OAAI,uBAAuB,IAAI,MAAM,KAAK,CAAE;GAE5C,MAAM,OAAO,KAAK,UAAU,MAAM,KAAK;AACvC,OAAI,MAAM,aAAa,CACrB,aAAY,MAAM,SAAS,sBAAsB;YACxC,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,EAErD;QAAI,MAAM,SAAS,YACjB,SAAQ,KAAK,KAAK;;;SAIlB;AAGR,QAAO;;;;;;;;;;;;;;;;;;;;;;;;;;AA+BT,MAAM,oBAAoB;;;;;;;;;;;;;AAc1B,SAAS,6BAA6B,UAA2B;CAC/D,MAAM,aAAa,UAAU,SAAS;CACtC,MAAM,OAAO,SAAS;AAGtB,KAAI,KAAK,WAAW,WAAW,IAAI,eAAe,IAChD,QAAO;AAIT,KAAI,WAAW,WAAW,KAAK,EAAE;EAC/B,MAAM,MAAM,WAAW,MAAM,KAAK,OAAO,CAAC,QAAQ,OAAO,GAAG;AAE5D,OADc,MAAM,IAAI,MAAM,IAAI,CAAC,SAAS,OAC9B,EAAG,QAAO;;AAK1B,KAAI,WAAW,KAAK,YAAY,OAAO,CAAC,CACtC,QAAO;AAGT,QAAO;;AAGT,eAAsB,aACpB,IACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAC1B,gBAAgB;EAChB,eAAe;EACf,cAAc;EACf;CAED,MAAM,eAA2F,EAAE;CAGnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAIvG,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;;CAI5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAMtF;EACE,MAAM,mBAAmB;EACzB,MAAM,mBAAmB,GAAG,QAAQ;;;MAGlC;EACF,MAAM,iBAAiB,GAAG,QAAQ;;;MAGhC;EACF,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAC3C,MAAM,WAAW,SAAS,QAAQ;GAClC,MAAM,IAAI,iBAAiB,KAAK,SAAS;AACzC,OAAI,CAAC,EAAG;GACR,MAAM,GAAG,KAAK,MAAM,SAAS;GAC7B,MAAM,OAAO,YAAY,IAAI,GAAG,KAAK,IAAI;GAEzC,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GACjC,MAAM,KAAK,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;GACrD,MAAM,OAAO,WAAW,KAAK;AAC7B,MAAG,kBAAkB;AACnB,qBAAiB,IAAI,IAAI,WAAW,SAAS,WAAW,eAAe,GAAG,GAAG,MAAM,MAAM,UAAU;AACnG,mBAAe,IAAI,MAAM,IAAI,WAAW,eAAe,SAAS,WAAW,GAAG,EAAE;KAChF,EAAE;;;AASR,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAMxF,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,0BAA0B;GAChC,MAAM,YAAY,KAAK,KAAK;GAC5B,MAAM,oBAAoB,GAAG,QAAQ;;;QAGnC;GACF,MAAM,kBAAkB,GAAG,QAAQ;;;QAGjC;AACF,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IACjD,MAAM,WAAW,SAAS,QAAQ;IAClC,MAAM,IAAI,wBAAwB,KAAK,SAAS;AAChD,QAAI,CAAC,EAAG;IACR,MAAM,GAAG,KAAK,MAAM,SAAS;IAC7B,MAAM,OAAO,YAAY,IAAI,GAAG,KAAK,IAAI;IAEzC,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IACjC,MAAM,KAAK,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IACrD,MAAM,OAAO,WAAW,KAAK;AAC7B,OAAG,kBAAkB;AACnB,uBAAkB,IAAI,IAAI,WAAW,SAAS,WAAW,eAAe,GAAG,GAAG,MAAM,MAAM,UAAU;AACpG,qBAAgB,IAAI,MAAM,IAAI,WAAW,eAAe,SAAS,WAAW,GAAG,EAAE;MACjF,EAAE;;;AAKR,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;GAGxD,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAChB,SAAS;IACT,UAAU;IACV,QAAQ;IACR,MAAM;IACP,CAAC;AAIJ,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;;;;AAMxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAE9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAG3C,MAFgB,UAAU,IAAI,WAAW,UAAU,SAAS,QAAQ,KAAK,EAE5D;GAEX,MAAM,QAAQ,GACX,QACC,4EACD,CACA,IAAI,WAAW,QAAQ;AAE1B,UAAO;AACP,UAAO,iBAAiB,MAAM;QAE9B,QAAO;;CAUX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAI5C,MAAM,eAAe,GAClB,QAAQ,+DAA+D,CACvE,IAAI,UAAU;CAEjB,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,OAAO,cAAc;EAE9B,MAAM,WAAW,IAAI,KAAK,SAAS,UAAU,GACzC,IAAI,KAAK,MAAM,GAAG,GAAkB,GACpC,IAAI;AACR,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,IAAI,KAAK;;AAI7B,KAAI,WAAW,SAAS,GAAG;EACzB,MAAM,kBAAkB,GAAG,QAAQ,sCAAsC;EACzE,MAAM,eAAe,GAAG,QACtB,8DACD;EACD,MAAM,aAAa,GAAG,QACpB,6DACD;AAED,KAAG,kBAAkB;AACnB,QAAK,MAAM,aAAa,YAAY;IAClC,MAAM,WAAW,GACd,QAAQ,iEAAiE,CACzE,IAAI,WAAW,UAAU;AAC5B,SAAK,MAAM,EAAE,QAAQ,SACnB,iBAAgB,IAAI,GAAG;AAEzB,iBAAa,IAAI,WAAW,UAAU;AACtC,eAAW,IAAI,WAAW,UAAU;;IAEtC,EAAE;;AAGN,QAAO;;;;;;AAWT,SAAS,mBAAkC;AACzC,QAAO,IAAI,SAAS,YAAY,aAAa,QAAQ,CAAC;;;;;;;;AASxD,eAAsB,SACpB,IACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAC1B,gBAAgB;EAChB,eAAe;EACf,cAAc;EACf;AAED,MAAK,MAAM,WAAW,UAAU;AAE9B,QAAM,kBAAkB;EAExB,MAAM,IAAI,MAAM,aAAa,IAAI,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzF,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ;;;;;;;;;;;;;;AAwBtD,eAAsB,YACpB,IACA,WACA,YAAY,IACZ,YACsB;CAEtB,MAAM,EAAE,mBAAmB,uBAAuB,MAAM,OAAO;CAE/D,MAAM,aAAa,CAAC,oBAAoB;CACxC,MAAM,SAA8B,EAAE;AAEtC,KAAI,cAAc,QAAW;AAC3B,aAAW,KAAK,iBAAiB;AACjC,SAAO,KAAK,UAAU;;CAGxB,MAAM,QAAQ,WAAW,WAAW,KAAK,QAAQ;CAEjD,MAAM,OAAO,GACV,QAAQ,sCAAsC,MAAM,cAAc,CAClE,IAAI,GAAG,OAAO;AAEjB,KAAI,KAAK,WAAW,EAClB,QAAO;EAAE,gBAAgB;EAAG,eAAe;EAAG;CAGhD,MAAM,aAAa,GAAG,QACpB,sDACD;CAED,IAAI,WAAW;CACf,MAAM,QAAQ,KAAK;AAGnB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,WAAW;EAC/C,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,UAAU;EAG1C,MAAM,aAAkD,EAAE;AAC1D,OAAK,MAAM,OAAO,OAAO;GAEvB,MAAM,OAAO,mBADD,MAAM,kBAAkB,IAAI,KAAK,CACT;AACpC,cAAW,KAAK;IAAE,IAAI,IAAI;IAAI;IAAM,CAAC;;AAIvC,KAAG,kBAAkB;AACnB,QAAK,MAAM,EAAE,IAAI,UAAU,WACzB,YAAW,IAAI,MAAM,GAAG;IAE1B,EAAE;AAEJ,cAAY,WAAW;AACvB,eAAa,UAAU,MAAM;;AAG/B,QAAO;EAAE,gBAAgB;EAAU,eAAe;EAAG"}

package/dist/{indexer-backend-DQO-FqAI.mjs → indexer-backend-CIMXedqk.mjs}

@@ -1,5 +1,5 @@
 import { t as chunkMarkdown } from "./chunker-CbnBe0s0.mjs";
-import { t as detectTier } from "./indexer-CKQcgKsz.mjs";
+import { t as detectTier } from "./indexer-CMPOiY1r.mjs";
 import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
 import { homedir } from "node:os";
 import { basename, join, normalize, relative } from "node:path";
@@ -223,7 +223,9 @@ async function indexProjectWithBackend(backend, projectId, rootPath, claudeNotes
 				updatedAt,
 				embedding: null
 			};
-			await backend.insertChunks([titleChunk]);
+			try {
+				await backend.insertChunks([titleChunk]);
+			} catch {}
 		}
 	}
 	if (!isPathTooBroadForContentScan(rootPath)) for (const absPath of walkContentFiles(rootPath)) filesToIndex.push({
@@ -258,7 +260,9 @@ async function indexProjectWithBackend(backend, projectId, rootPath, claudeNotes
 					updatedAt,
 					embedding: null
 				};
-				await backend.insertChunks([titleChunk]);
+				try {
+					await backend.insertChunks([titleChunk]);
+				} catch {}
 			}
 		}
 		if (claudeNotesDir.endsWith("/Notes")) {
@@ -291,12 +295,25 @@ async function indexProjectWithBackend(backend, projectId, rootPath, claudeNotes
 		}
 		filesSinceYield++;
 		const relPath = relative(rootBase, absPath);
-		if (await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier)) {
-			const ids = await backend.getChunkIds(projectId, relPath);
-			result.filesProcessed++;
-			result.chunksCreated += ids.length;
-		} else result.filesSkipped++;
+		try {
+			if (await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier)) {
+				const ids = await backend.getChunkIds(projectId, relPath);
+				result.filesProcessed++;
+				result.chunksCreated += ids.length;
+			} else result.filesSkipped++;
+		} catch {
+			result.filesSkipped++;
+		}
+	}
+	const livePaths = /* @__PURE__ */ new Set();
+	for (const { absPath, rootBase } of filesToIndex) livePaths.add(relative(rootBase, absPath));
+	const dbChunkPaths = await backend.getDistinctChunkPaths(projectId);
+	const stalePaths = [];
+	for (const p of dbChunkPaths) {
+		const basePath = p.endsWith("::title") ? p.slice(0, -7) : p;
+		if (!livePaths.has(basePath)) stalePaths.push(p);
 	}
+	if (stalePaths.length > 0) await backend.deletePaths(projectId, stalePaths);
 	return result;
 }
 const EMBED_BATCH_SIZE = 50;
@@ -358,4 +375,4 @@ async function indexAllWithBackend(backend, registryDb) {
 
 //#endregion
 export { embedChunksWithBackend, indexAllWithBackend };
-//# sourceMappingURL=indexer-backend-DQO-FqAI.mjs.map
+//# sourceMappingURL=indexer-backend-CIMXedqk.mjs.map

package/dist/indexer-backend-CIMXedqk.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"indexer-backend-CIMXedqk.mjs","names":[],"sources":["../src/memory/indexer-backend.ts"],"sourcesContent":["/**\n * Backend-aware indexer for PAI federation memory.\n *\n * This module provides the same functionality as indexer.ts but writes\n * through the 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 indexer.ts directly (which is faster for SQLite\n * due to synchronous transactions).\n */\n\nimport { createHash } from \"node:crypto\";\nimport { readFileSync, statSync, readdirSync, existsSync } from \"node:fs\";\nimport { join, relative, basename, normalize } from \"node:path\";\n\n// ---------------------------------------------------------------------------\n// Session title parsing\n// ---------------------------------------------------------------------------\n\nconst SESSION_TITLE_RE = /^(\\d{4})\\s*-\\s*(\\d{4}-\\d{2}-\\d{2})\\s*-\\s*(.+)\\.md$/;\n\n/**\n * Parse a session title from a Notes filename.\n * Format: \"NNNN - YYYY-MM-DD - Descriptive Title.md\"\n * Returns a synthetic chunk text like \"Session #0086 2026-02-23: Pai Daemon Background Service\"\n * or null if the filename doesn't match the expected pattern.\n */\nexport function parseSessionTitleChunk(fileName: string): string | null {\n  const m = SESSION_TITLE_RE.exec(fileName);\n  if (!m) return null;\n  const [, num, date, title] = m;\n  return `Session #${num} ${date}: ${title}`;\n}\nimport { homedir } from \"node:os\";\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow } from \"../storage/interface.js\";\nimport type { IndexResult } from \"./indexer.js\";\nimport { chunkMarkdown } from \"./chunker.js\";\nimport { detectTier } from \"./indexer.js\";\n\n// ---------------------------------------------------------------------------\n// Constants (mirrored from indexer.ts)\n// ---------------------------------------------------------------------------\n\nconst MAX_FILES_PER_PROJECT = 5_000;\nconst MAX_WALK_DEPTH = 6;\nconst INDEX_YIELD_EVERY = 10;\n\n/**\n * Directories to ALWAYS skip, at any depth, during any directory walk.\n * These are build artifacts, dependency trees, and VCS internals that\n * should never be indexed regardless of where they appear in the tree.\n */\nconst ALWAYS_SKIP_DIRS = new Set([\n  // Version control\n  \".git\",\n  // Dependency directories (any language)\n  \"node_modules\",\n  \"vendor\",\n  \"Pods\",              // CocoaPods (iOS/macOS)\n  // Build / compile output\n  \"dist\",\n  \"build\",\n  \"out\",\n  \"DerivedData\",       // Xcode\n  \".next\",             // Next.js\n  // Python virtual environments and caches\n  \".venv\",\n  \"venv\",\n  \"__pycache__\",\n  // General caches\n  \".cache\",\n  \".bun\",\n]);\n\nconst ROOT_SCAN_SKIP_DIRS = new Set([\n  \"memory\", \"Notes\", \".claude\", \".DS_Store\",\n  ...ALWAYS_SKIP_DIRS,\n]);\n\nconst CONTENT_SCAN_SKIP_DIRS = new Set([\n  \"Library\", \"Applications\", \"Music\", \"Movies\", \"Pictures\", \"Desktop\",\n  \"Downloads\", \"Public\", \"coverage\",\n  ...ALWAYS_SKIP_DIRS,\n]);\n\n// ---------------------------------------------------------------------------\n// Helpers (same logic as indexer.ts)\n// ---------------------------------------------------------------------------\n\nfunction sha256File(content: string): string {\n  return createHash(\"sha256\").update(content).digest(\"hex\");\n}\n\nfunction chunkId(\n  projectId: number,\n  path: string,\n  chunkIndex: number,\n  startLine: number,\n  endLine: number,\n): string {\n  return createHash(\"sha256\")\n    .update(`${projectId}:${path}:${chunkIndex}:${startLine}:${endLine}`)\n    .digest(\"hex\");\n}\n\nfunction walkMdFiles(\n  dir: string,\n  acc?: string[],\n  cap = MAX_FILES_PER_PROJECT,\n  depth = 0,\n): string[] {\n  const results = acc ?? [];\n  if (!existsSync(dir)) return results;\n  if (results.length >= cap) return results;\n  if (depth > MAX_WALK_DEPTH) return results;\n  try {\n    for (const entry of readdirSync(dir, { withFileTypes: true })) {\n      if (results.length >= cap) break;\n      if (entry.isSymbolicLink()) continue;\n      // Skip known junk directories at every recursion depth\n      if (ALWAYS_SKIP_DIRS.has(entry.name)) continue;\n      const full = join(dir, entry.name);\n      if (entry.isDirectory()) {\n        walkMdFiles(full, results, cap, depth + 1);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        results.push(full);\n      }\n    }\n  } catch { /* skip unreadable */ }\n  return results;\n}\n\nfunction walkContentFiles(rootPath: string): string[] {\n  if (!existsSync(rootPath)) return [];\n  const results: string[] = [];\n  try {\n    for (const entry of readdirSync(rootPath, { withFileTypes: true })) {\n      if (results.length >= MAX_FILES_PER_PROJECT) break;\n      if (entry.isSymbolicLink()) continue;\n      if (ROOT_SCAN_SKIP_DIRS.has(entry.name)) continue;\n      if (CONTENT_SCAN_SKIP_DIRS.has(entry.name)) continue;\n      const full = join(rootPath, entry.name);\n      if (entry.isDirectory()) {\n        walkMdFiles(full, results, MAX_FILES_PER_PROJECT);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        if (entry.name !== \"MEMORY.md\") results.push(full);\n      }\n    }\n  } catch { /* skip */ }\n  return results;\n}\n\nfunction isPathTooBroadForContentScan(rootPath: string): boolean {\n  const normalized = normalize(rootPath);\n  const home = homedir();\n  if (home.startsWith(normalized) || normalized === \"/\") return true;\n  if (normalized.startsWith(home)) {\n    const rel = normalized.slice(home.length).replace(/^\\//, \"\");\n    const depth = rel ? rel.split(\"/\").length : 0;\n    if (depth === 0) return true;\n  }\n  if (existsSync(join(normalized, \".git\"))) return true;\n  return false;\n}\n\nfunction yieldToEventLoop(): Promise<void> {\n  return new Promise((resolve) => setImmediate(resolve));\n}\n\n// ---------------------------------------------------------------------------\n// 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: parse titles from Notes filenames and insert\n  // as high-signal chunks so session names are searchable via BM25 and embeddings.\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 = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      const titleChunk: import(\"../storage/interface.js\").ChunkRow = {\n        id, projectId, source: \"notes\", tier: \"session\",\n        path: syntheticPath, startLine: 0, endLine: 0,\n        hash, text, updatedAt, embedding: null,\n      };\n      try {\n        await backend.insertChunks([titleChunk]);\n      } catch {\n        // Skip title chunks that cause backend errors\n      }\n    }\n  }\n\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        const titleChunk: import(\"../storage/interface.js\").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        // Count chunks — we know we just inserted them, count from the chunk IDs\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  // ---------------------------------------------------------------------------\n  // Prune stale paths: remove DB entries for files that no longer exist on disk.\n  // This handles renames, moves, and deletions — the indexer only adds/updates,\n  // so without pruning, old paths accumulate forever.\n  // ---------------------------------------------------------------------------\n\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 = 10;\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): 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  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 } = batch[j];\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    }\n\n    embedded += batch.length;\n    process.stderr.write(\n      `[pai-daemon] Embedded ${embedded}/${total} chunks\\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":";;;;;;;;;;;;;;;;;;AAmBA,MAAM,mBAAmB;;;;;;;AAQzB,SAAgB,uBAAuB,UAAiC;CACtE,MAAM,IAAI,iBAAiB,KAAK,SAAS;AACzC,KAAI,CAAC,EAAG,QAAO;CACf,MAAM,GAAG,KAAK,MAAM,SAAS;AAC7B,QAAO,YAAY,IAAI,GAAG,KAAK,IAAI;;AAarC,MAAM,wBAAwB;AAC9B,MAAM,iBAAiB;AACvB,MAAM,oBAAoB;;;;;;AAO1B,MAAM,mBAAmB,IAAI,IAAI;CAE/B;CAEA;CACA;CACA;CAEA;CACA;CACA;CACA;CACA;CAEA;CACA;CACA;CAEA;CACA;CACD,CAAC;AAEF,MAAM,sBAAsB,IAAI,IAAI;CAClC;CAAU;CAAS;CAAW;CAC9B,GAAG;CACJ,CAAC;AAEF,MAAM,yBAAyB,IAAI,IAAI;CACrC;CAAW;CAAgB;CAAS;CAAU;CAAY;CAC1D;CAAa;CAAU;CACvB,GAAG;CACJ,CAAC;AAMF,SAAS,WAAW,SAAyB;AAC3C,QAAO,WAAW,SAAS,CAAC,OAAO,QAAQ,CAAC,OAAO,MAAM;;AAG3D,SAAS,QACP,WACA,MACA,YACA,WACA,SACQ;AACR,QAAO,WAAW,SAAS,CACxB,OAAO,GAAG,UAAU,GAAG,KAAK,GAAG,WAAW,GAAG,UAAU,GAAG,UAAU,CACpE,OAAO,MAAM;;AAGlB,SAAS,YACP,KACA,KACA,MAAM,uBACN,QAAQ,GACE;CACV,MAAM,UAAU,OAAO,EAAE;AACzB,KAAI,CAAC,WAAW,IAAI,CAAE,QAAO;AAC7B,KAAI,QAAQ,UAAU,IAAK,QAAO;AAClC,KAAI,QAAQ,eAAgB,QAAO;AACnC,KAAI;AACF,OAAK,MAAM,SAAS,YAAY,KAAK,EAAE,eAAe,MAAM,CAAC,EAAE;AAC7D,OAAI,QAAQ,UAAU,IAAK;AAC3B,OAAI,MAAM,gBAAgB,CAAE;AAE5B,OAAI,iBAAiB,IAAI,MAAM,KAAK,CAAE;GACtC,MAAM,OAAO,KAAK,KAAK,MAAM,KAAK;AAClC,OAAI,MAAM,aAAa,CACrB,aAAY,MAAM,SAAS,KAAK,QAAQ,EAAE;YACjC,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,CACrD,SAAQ,KAAK,KAAK;;SAGhB;AACR,QAAO;;AAGT,SAAS,iBAAiB,UAA4B;AACpD,KAAI,CAAC,WAAW,SAAS,CAAE,QAAO,EAAE;CACpC,MAAM,UAAoB,EAAE;AAC5B,KAAI;AACF,OAAK,MAAM,SAAS,YAAY,UAAU,EAAE,eAAe,MAAM,CAAC,EAAE;AAClE,OAAI,QAAQ,UAAU,sBAAuB;AAC7C,OAAI,MAAM,gBAAgB,CAAE;AAC5B,OAAI,oBAAoB,IAAI,MAAM,KAAK,CAAE;AACzC,OAAI,uBAAuB,IAAI,MAAM,KAAK,CAAE;GAC5C,MAAM,OAAO,KAAK,UAAU,MAAM,KAAK;AACvC,OAAI,MAAM,aAAa,CACrB,aAAY,MAAM,SAAS,sBAAsB;YACxC,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,EACrD;QAAI,MAAM,SAAS,YAAa,SAAQ,KAAK,KAAK;;;SAGhD;AACR,QAAO;;AAGT,SAAS,6BAA6B,UAA2B;CAC/D,MAAM,aAAa,UAAU,SAAS;CACtC,MAAM,OAAO,SAAS;AACtB,KAAI,KAAK,WAAW,WAAW,IAAI,eAAe,IAAK,QAAO;AAC9D,KAAI,WAAW,WAAW,KAAK,EAAE;EAC/B,MAAM,MAAM,WAAW,MAAM,KAAK,OAAO,CAAC,QAAQ,OAAO,GAAG;AAE5D,OADc,MAAM,IAAI,MAAM,IAAI,CAAC,SAAS,OAC9B,EAAG,QAAO;;AAE1B,KAAI,WAAW,KAAK,YAAY,OAAO,CAAC,CAAE,QAAO;AACjD,QAAO;;AAGT,SAAS,mBAAkC;AACzC,QAAO,IAAI,SAAS,YAAY,aAAa,QAAQ,CAAC;;;;;;AAWxD,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;CAKtF;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,aAAyD;IAC7D,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,aAAyD;KAC7D,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;IAEX,MAAM,MAAM,MAAM,QAAQ,YAAY,WAAW,QAAQ;AACzD,WAAO;AACP,WAAO,iBAAiB,IAAI;SAE5B,QAAO;UAEH;AAEN,UAAO;;;CAUX,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,YACiB;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;AAEf,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,SAAS,MAAM;AAG3B,QAAK,WAAW,KAAK,sBAAsB,EACzC,OAAM,kBAAkB;GAI1B,MAAM,OAAO,mBADD,MAAM,kBAAkB,KAAK,CACL;AACpC,SAAM,QAAQ,gBAAgB,IAAI,KAAK;;AAGzC,cAAY,MAAM;AAClB,UAAQ,OAAO,MACb,yBAAyB,SAAS,GAAG,MAAM,WAC5C;;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/{postgres-CIxeqf_n.mjs → postgres-FXrHDPcE.mjs}

@@ -90,7 +90,9 @@ var PostgresBackend = class {
 		const client = await this.pool.connect();
 		try {
 			await client.query("BEGIN");
-			for (const c of chunks) await client.query(`INSERT INTO pai_chunks
+			for (const c of chunks) {
+				const safeText = c.text.replace(/\0/g, "");
+				await client.query(`INSERT INTO pai_chunks
              (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at, fts_vector)
            VALUES
              ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10,
@@ -106,17 +108,38 @@ var PostgresBackend = class {
              text       = EXCLUDED.text,
              updated_at = EXCLUDED.updated_at,
              fts_vector = EXCLUDED.fts_vector`, [
-				c.id,
-				c.projectId,
-				c.source,
-				c.tier,
-				c.path,
-				c.startLine,
-				c.endLine,
-				c.hash,
-				c.text,
-				c.updatedAt
-			]);
+					c.id,
+					c.projectId,
+					c.source,
+					c.tier,
+					c.path,
+					c.startLine,
+					c.endLine,
+					c.hash,
+					safeText,
+					c.updatedAt
+				]);
+			}
+			await client.query("COMMIT");
+		} catch (e) {
+			await client.query("ROLLBACK");
+			throw e;
+		} finally {
+			client.release();
+		}
+	}
+	async getDistinctChunkPaths(projectId) {
+		return (await this.pool.query("SELECT DISTINCT path FROM pai_chunks WHERE project_id = $1", [projectId])).rows.map((r) => r.path);
+	}
+	async deletePaths(projectId, paths) {
+		if (paths.length === 0) return;
+		const client = await this.pool.connect();
+		try {
+			await client.query("BEGIN");
+			for (const path of paths) {
+				await client.query("DELETE FROM pai_chunks WHERE project_id = $1 AND path = $2", [projectId, path]);
+				await client.query("DELETE FROM pai_files WHERE project_id = $1 AND path = $2", [projectId, path]);
+			}
 			await client.query("COMMIT");
 		} catch (e) {
 			await client.query("ROLLBACK");
@@ -332,4 +355,4 @@ function buildPgTsQuery(query) {
 
 //#endregion
 export { PostgresBackend };
-//# sourceMappingURL=postgres-CIxeqf_n.mjs.map
+//# sourceMappingURL=postgres-FXrHDPcE.mjs.map

package/dist/postgres-FXrHDPcE.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"postgres-FXrHDPcE.mjs","names":[],"sources":["../src/storage/postgres.ts"],"sourcesContent":["/**\n * PostgresBackend — implements StorageBackend using PostgreSQL + pgvector.\n *\n * Vector similarity: pgvector's <=> cosine distance operator\n * Full-text search:  PostgreSQL tsvector/tsquery (replaces SQLite FTS5)\n * Connection pooling: node-postgres Pool\n *\n * Schema is initialized via docker/init.sql.\n * This module only handles runtime queries — schema creation is external.\n */\n\nimport pg from \"pg\";\nimport type { Pool, PoolClient } from \"pg\";\nimport type { StorageBackend, ChunkRow, FileRow, FederationStats } from \"./interface.js\";\nimport type { SearchResult, SearchOptions } from \"../memory/search.js\";\nimport { buildFtsQuery } from \"../memory/search.js\";\n\nconst { Pool: PgPool } = pg;\n\n// ---------------------------------------------------------------------------\n// Postgres config\n// ---------------------------------------------------------------------------\n\nexport interface PostgresConfig {\n  connectionString?: string;\n  host?: string;\n  port?: number;\n  database?: string;\n  user?: string;\n  password?: string;\n  /** Maximum pool connections. Default 5 */\n  maxConnections?: number;\n  /** Connection timeout in ms. Default 5000 */\n  connectionTimeoutMs?: number;\n}\n\n// ---------------------------------------------------------------------------\n// Implementation\n// ---------------------------------------------------------------------------\n\nexport class PostgresBackend implements StorageBackend {\n  readonly backendType = \"postgres\" as const;\n\n  private pool: Pool;\n\n  constructor(config: PostgresConfig) {\n    const connStr =\n      config.connectionString ??\n      `postgresql://${config.user ?? \"pai\"}:${config.password ?? \"pai\"}@${config.host ?? \"localhost\"}:${config.port ?? 5432}/${config.database ?? \"pai\"}`;\n\n    this.pool = new PgPool({\n      connectionString: connStr,\n      max: config.maxConnections ?? 5,\n      connectionTimeoutMillis: config.connectionTimeoutMs ?? 5000,\n      idleTimeoutMillis: 30_000,\n    });\n\n    // Log pool errors so they don't crash the process silently\n    this.pool.on(\"error\", (err) => {\n      process.stderr.write(`[pai-postgres] Pool error: ${err.message}\\n`);\n    });\n  }\n\n  // -------------------------------------------------------------------------\n  // Lifecycle\n  // -------------------------------------------------------------------------\n\n  async close(): Promise<void> {\n    await this.pool.end();\n  }\n\n  async getStats(): Promise<FederationStats> {\n    const client = await this.pool.connect();\n    try {\n      const filesResult = await client.query<{ n: string }>(\n        \"SELECT COUNT(*)::text AS n FROM pai_files\"\n      );\n      const chunksResult = await client.query<{ n: string }>(\n        \"SELECT COUNT(*)::text AS n FROM pai_chunks\"\n      );\n      return {\n        files: parseInt(filesResult.rows[0]?.n ?? \"0\", 10),\n        chunks: parseInt(chunksResult.rows[0]?.n ?? \"0\", 10),\n      };\n    } finally {\n      client.release();\n    }\n  }\n\n  /**\n   * Test the connection by running a trivial query.\n   * Returns null on success, error message on failure.\n   */\n  async testConnection(): Promise<string | null> {\n    let client: PoolClient | null = null;\n    try {\n      client = await this.pool.connect();\n      await client.query(\"SELECT 1\");\n      return null;\n    } catch (e) {\n      return e instanceof Error ? e.message : String(e);\n    } finally {\n      client?.release();\n    }\n  }\n\n  // -------------------------------------------------------------------------\n  // File tracking\n  // -------------------------------------------------------------------------\n\n  async getFileHash(projectId: number, path: string): Promise<string | undefined> {\n    const result = await this.pool.query<{ hash: string }>(\n      \"SELECT hash FROM pai_files WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n    return result.rows[0]?.hash;\n  }\n\n  async upsertFile(file: FileRow): Promise<void> {\n    await this.pool.query(\n      `INSERT INTO pai_files (project_id, path, source, tier, hash, mtime, size)\n       VALUES ($1, $2, $3, $4, $5, $6, $7)\n       ON CONFLICT (project_id, path) DO UPDATE SET\n         source = EXCLUDED.source,\n         tier   = EXCLUDED.tier,\n         hash   = EXCLUDED.hash,\n         mtime  = EXCLUDED.mtime,\n         size   = EXCLUDED.size`,\n      [file.projectId, file.path, file.source, file.tier, file.hash, file.mtime, file.size]\n    );\n  }\n\n  // -------------------------------------------------------------------------\n  // Chunk management\n  // -------------------------------------------------------------------------\n\n  async getChunkIds(projectId: number, path: string): Promise<string[]> {\n    const result = await this.pool.query<{ id: string }>(\n      \"SELECT id FROM pai_chunks WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n    return result.rows.map((r) => r.id);\n  }\n\n  async deleteChunksForFile(projectId: number, path: string): Promise<void> {\n    // Foreign key CASCADE handles pai_chunks deletion automatically\n    // but we don't have FK to pai_chunks from pai_files, so delete explicitly\n    await this.pool.query(\n      \"DELETE FROM pai_chunks WHERE project_id = $1 AND path = $2\",\n      [projectId, path]\n    );\n  }\n\n  async insertChunks(chunks: ChunkRow[]): Promise<void> {\n    if (chunks.length === 0) return;\n\n    const client = await this.pool.connect();\n    try {\n      await client.query(\"BEGIN\");\n\n      for (const c of chunks) {\n        // Strip null bytes — Postgres rejects \\0 in text columns\n        const safeText = c.text.replace(/\\0/g, \"\");\n\n        // embedding is null at insert time; updated separately via updateEmbedding()\n        await client.query(\n          `INSERT INTO pai_chunks\n             (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at, fts_vector)\n           VALUES\n             ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10,\n              to_tsvector('simple', $9))\n           ON CONFLICT (id) DO UPDATE SET\n             project_id = EXCLUDED.project_id,\n             source     = EXCLUDED.source,\n             tier       = EXCLUDED.tier,\n             path       = EXCLUDED.path,\n             start_line = EXCLUDED.start_line,\n             end_line   = EXCLUDED.end_line,\n             hash       = EXCLUDED.hash,\n             text       = EXCLUDED.text,\n             updated_at = EXCLUDED.updated_at,\n             fts_vector = EXCLUDED.fts_vector`,\n          [\n            c.id,\n            c.projectId,\n            c.source,\n            c.tier,\n            c.path,\n            c.startLine,\n            c.endLine,\n            c.hash,\n            safeText,\n            c.updatedAt,\n          ]\n        );\n      }\n\n      await client.query(\"COMMIT\");\n    } catch (e) {\n      await client.query(\"ROLLBACK\");\n      throw e;\n    } finally {\n      client.release();\n    }\n  }\n\n  async getDistinctChunkPaths(projectId: number): Promise<string[]> {\n    const result = await this.pool.query<{ path: string }>(\n      \"SELECT DISTINCT path FROM pai_chunks WHERE project_id = $1\",\n      [projectId]\n    );\n    return result.rows.map((r) => r.path);\n  }\n\n  async deletePaths(projectId: number, paths: string[]): Promise<void> {\n    if (paths.length === 0) return;\n    const client = await this.pool.connect();\n    try {\n      await client.query(\"BEGIN\");\n      for (const path of paths) {\n        await client.query(\n          \"DELETE FROM pai_chunks WHERE project_id = $1 AND path = $2\",\n          [projectId, path]\n        );\n        await client.query(\n          \"DELETE FROM pai_files WHERE project_id = $1 AND path = $2\",\n          [projectId, path]\n        );\n      }\n      await client.query(\"COMMIT\");\n    } catch (e) {\n      await client.query(\"ROLLBACK\");\n      throw e;\n    } finally {\n      client.release();\n    }\n  }\n\n  async getUnembeddedChunkIds(projectId?: number): Promise<Array<{ id: string; text: string }>> {\n    if (projectId !== undefined) {\n      const result = await this.pool.query<{ id: string; text: string }>(\n        \"SELECT id, text FROM pai_chunks WHERE embedding IS NULL AND project_id = $1 ORDER BY id\",\n        [projectId]\n      );\n      return result.rows;\n    }\n    const result = await this.pool.query<{ id: string; text: string }>(\n      \"SELECT id, text FROM pai_chunks WHERE embedding IS NULL ORDER BY id\"\n    );\n    return result.rows;\n  }\n\n  async updateEmbedding(chunkId: string, embedding: Buffer): Promise<void> {\n    // Deserialize the Buffer (Float32Array LE bytes) to a number[] for pgvector\n    const vec = bufferToVector(embedding);\n    const vecStr = \"[\" + vec.join(\",\") + \"]\";\n    await this.pool.query(\n      \"UPDATE pai_chunks SET embedding = $1::vector WHERE id = $2\",\n      [vecStr, chunkId]\n    );\n  }\n\n  // -------------------------------------------------------------------------\n  // Search — keyword (tsvector/tsquery)\n  // -------------------------------------------------------------------------\n\n  async searchKeyword(query: string, opts?: SearchOptions): Promise<SearchResult[]> {\n    const maxResults = opts?.maxResults ?? 10;\n\n    // Build tsquery from the same token logic as buildFtsQuery, but for Postgres\n    const tsQuery = buildPgTsQuery(query);\n    if (!tsQuery) return [];\n\n    // Use 'simple' dictionary: preserves tokens as-is, no language-specific\n    // stemming. Works reliably with any language (German, French, etc.).\n    const conditions: string[] = [\"fts_vector @@ to_tsquery('simple', $1)\"];\n    const params: (string | number)[] = [tsQuery];\n    let paramIdx = 2;\n\n    if (opts?.projectIds && opts.projectIds.length > 0) {\n      const placeholders = opts.projectIds.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`project_id IN (${placeholders})`);\n      params.push(...opts.projectIds);\n    }\n\n    if (opts?.sources && opts.sources.length > 0) {\n      const placeholders = opts.sources.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`source IN (${placeholders})`);\n      params.push(...opts.sources);\n    }\n\n    if (opts?.tiers && opts.tiers.length > 0) {\n      const placeholders = opts.tiers.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`tier IN (${placeholders})`);\n      params.push(...opts.tiers);\n    }\n\n    params.push(maxResults);\n    const limitParam = `$${paramIdx}`;\n\n    const sql = `\n      SELECT\n        project_id,\n        path,\n        start_line,\n        end_line,\n        text AS snippet,\n        tier,\n        source,\n        ts_rank(fts_vector, to_tsquery('simple', $1)) AS rank_score\n      FROM pai_chunks\n      WHERE ${conditions.join(\" AND \")}\n      ORDER BY rank_score DESC\n      LIMIT ${limitParam}\n    `;\n\n    try {\n      const result = await this.pool.query<{\n        project_id: number;\n        path: string;\n        start_line: number;\n        end_line: number;\n        snippet: string;\n        tier: string;\n        source: string;\n        rank_score: number;\n      }>(sql, params);\n\n      return result.rows.map((row) => ({\n        projectId: row.project_id,\n        path: row.path,\n        startLine: row.start_line,\n        endLine: row.end_line,\n        snippet: row.snippet,\n        score: row.rank_score,\n        tier: row.tier,\n        source: row.source,\n      }));\n    } catch (e) {\n      process.stderr.write(`[pai-postgres] searchKeyword error: ${e}\\n`);\n      return [];\n    }\n  }\n\n  // -------------------------------------------------------------------------\n  // Search — semantic (pgvector cosine distance)\n  // -------------------------------------------------------------------------\n\n  async searchSemantic(queryEmbedding: Float32Array, opts?: SearchOptions): Promise<SearchResult[]> {\n    const maxResults = opts?.maxResults ?? 10;\n\n    const conditions: string[] = [\"embedding IS NOT NULL\"];\n    const params: (string | number | string)[] = [];\n    let paramIdx = 1;\n\n    // pgvector vector literal\n    const vecStr = \"[\" + Array.from(queryEmbedding).join(\",\") + \"]\";\n    params.push(vecStr);\n    const vecParam = `$${paramIdx++}`;\n\n    if (opts?.projectIds && opts.projectIds.length > 0) {\n      const placeholders = opts.projectIds.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`project_id IN (${placeholders})`);\n      params.push(...opts.projectIds);\n    }\n\n    if (opts?.sources && opts.sources.length > 0) {\n      const placeholders = opts.sources.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`source IN (${placeholders})`);\n      params.push(...opts.sources);\n    }\n\n    if (opts?.tiers && opts.tiers.length > 0) {\n      const placeholders = opts.tiers.map(() => `$${paramIdx++}`).join(\", \");\n      conditions.push(`tier IN (${placeholders})`);\n      params.push(...opts.tiers);\n    }\n\n    params.push(maxResults);\n    const limitParam = `$${paramIdx}`;\n\n    // <=> is cosine distance; 1 - distance = cosine similarity\n    const sql = `\n      SELECT\n        project_id,\n        path,\n        start_line,\n        end_line,\n        text AS snippet,\n        tier,\n        source,\n        1 - (embedding <=> ${vecParam}::vector) AS cosine_similarity\n      FROM pai_chunks\n      WHERE ${conditions.join(\" AND \")}\n      ORDER BY embedding <=> ${vecParam}::vector\n      LIMIT ${limitParam}\n    `;\n\n    try {\n      const result = await this.pool.query<{\n        project_id: number;\n        path: string;\n        start_line: number;\n        end_line: number;\n        snippet: string;\n        tier: string;\n        source: string;\n        cosine_similarity: number;\n      }>(sql, params);\n\n      const minScore = opts?.minScore ?? -Infinity;\n\n      return result.rows\n        .map((row) => ({\n          projectId: row.project_id,\n          path: row.path,\n          startLine: row.start_line,\n          endLine: row.end_line,\n          snippet: row.snippet,\n          score: row.cosine_similarity,\n          tier: row.tier,\n          source: row.source,\n        }))\n        .filter((r) => r.score >= minScore);\n    } catch (e) {\n      process.stderr.write(`[pai-postgres] searchSemantic error: ${e}\\n`);\n      return [];\n    }\n  }\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\n/**\n * Convert a Buffer of Float32 LE bytes (as stored in SQLite) to number[].\n */\nfunction bufferToVector(buf: Buffer): number[] {\n  const floats: number[] = [];\n  for (let i = 0; i < buf.length; i += 4) {\n    floats.push(buf.readFloatLE(i));\n  }\n  return floats;\n}\n\n/**\n * Convert a free-text query to a Postgres tsquery string.\n *\n * Uses OR (|) semantics so that a chunk matching ANY query term is returned,\n * ranked by ts_rank (which scores higher when more terms match).  AND (&)\n * semantics are too strict for multi-word queries because all terms rarely\n * co-occur in a single chunk.\n *\n * Example: \"Synchrotech interview follow-up Gilles\"\n *   → \"synchrotech | interview | follow | gilles\"\n *   → returns chunks containing any of these words, highest-matching first\n */\nfunction buildPgTsQuery(query: string): string {\n  const STOP_WORDS = new Set([\n    \"a\", \"an\", \"and\", \"are\", \"as\", \"at\", \"be\", \"been\", \"but\", \"by\",\n    \"do\", \"for\", \"from\", \"has\", \"have\", \"he\", \"her\", \"him\", \"his\",\n    \"how\", \"i\", \"if\", \"in\", \"is\", \"it\", \"its\", \"me\", \"my\", \"not\",\n    \"of\", \"on\", \"or\", \"our\", \"out\", \"she\", \"so\", \"that\", \"the\",\n    \"their\", \"them\", \"they\", \"this\", \"to\", \"up\", \"us\", \"was\", \"we\",\n    \"were\", \"what\", \"when\", \"who\", \"will\", \"with\", \"you\", \"your\",\n  ]);\n\n  const tokens = query\n    .toLowerCase()\n    .split(/[\\s\\p{P}]+/u)\n    .filter(Boolean)\n    .filter((t) => t.length >= 2)\n    .filter((t) => !STOP_WORDS.has(t))\n    // Sanitize: strip tsquery special characters to prevent syntax errors\n    .map((t) => t.replace(/'/g, \"''\").replace(/[&|!():]/g, \"\"))\n    .filter(Boolean);\n\n  if (tokens.length === 0) {\n    // Fallback: sanitize the raw query and use it as a single term\n    const raw = query.replace(/[^a-z0-9]/gi, \" \").trim().split(/\\s+/).filter(Boolean).join(\" | \");\n    return raw || \"\";\n  }\n\n  // Use OR (|) so that chunks matching ANY term are returned.\n  // ts_rank naturally scores chunks higher when more terms match, so the\n  // most relevant results still bubble to the top.\n  return tokens.join(\" | \");\n}\n\n// Re-export buildFtsQuery so it is accessible without importing search.ts\nexport { buildPgTsQuery };\n"],"mappings":";;;;;;;;;;;;;AAiBA,MAAM,EAAE,MAAM,WAAW;AAuBzB,IAAa,kBAAb,MAAuD;CACrD,AAAS,cAAc;CAEvB,AAAQ;CAER,YAAY,QAAwB;AAKlC,OAAK,OAAO,IAAI,OAAO;GACrB,kBAJA,OAAO,oBACP,gBAAgB,OAAO,QAAQ,MAAM,GAAG,OAAO,YAAY,MAAM,GAAG,OAAO,QAAQ,YAAY,GAAG,OAAO,QAAQ,KAAK,GAAG,OAAO,YAAY;GAI5I,KAAK,OAAO,kBAAkB;GAC9B,yBAAyB,OAAO,uBAAuB;GACvD,mBAAmB;GACpB,CAAC;AAGF,OAAK,KAAK,GAAG,UAAU,QAAQ;AAC7B,WAAQ,OAAO,MAAM,8BAA8B,IAAI,QAAQ,IAAI;IACnE;;CAOJ,MAAM,QAAuB;AAC3B,QAAM,KAAK,KAAK,KAAK;;CAGvB,MAAM,WAAqC;EACzC,MAAM,SAAS,MAAM,KAAK,KAAK,SAAS;AACxC,MAAI;GACF,MAAM,cAAc,MAAM,OAAO,MAC/B,4CACD;GACD,MAAM,eAAe,MAAM,OAAO,MAChC,6CACD;AACD,UAAO;IACL,OAAO,SAAS,YAAY,KAAK,IAAI,KAAK,KAAK,GAAG;IAClD,QAAQ,SAAS,aAAa,KAAK,IAAI,KAAK,KAAK,GAAG;IACrD;YACO;AACR,UAAO,SAAS;;;;;;;CAQpB,MAAM,iBAAyC;EAC7C,IAAI,SAA4B;AAChC,MAAI;AACF,YAAS,MAAM,KAAK,KAAK,SAAS;AAClC,SAAM,OAAO,MAAM,WAAW;AAC9B,UAAO;WACA,GAAG;AACV,UAAO,aAAa,QAAQ,EAAE,UAAU,OAAO,EAAE;YACzC;AACR,WAAQ,SAAS;;;CAQrB,MAAM,YAAY,WAAmB,MAA2C;AAK9E,UAJe,MAAM,KAAK,KAAK,MAC7B,kEACA,CAAC,WAAW,KAAK,CAClB,EACa,KAAK,IAAI;;CAGzB,MAAM,WAAW,MAA8B;AAC7C,QAAM,KAAK,KAAK,MACd;;;;;;;kCAQA;GAAC,KAAK;GAAW,KAAK;GAAM,KAAK;GAAQ,KAAK;GAAM,KAAK;GAAM,KAAK;GAAO,KAAK;GAAK,CACtF;;CAOH,MAAM,YAAY,WAAmB,MAAiC;AAKpE,UAJe,MAAM,KAAK,KAAK,MAC7B,iEACA,CAAC,WAAW,KAAK,CAClB,EACa,KAAK,KAAK,MAAM,EAAE,GAAG;;CAGrC,MAAM,oBAAoB,WAAmB,MAA6B;AAGxE,QAAM,KAAK,KAAK,MACd,8DACA,CAAC,WAAW,KAAK,CAClB;;CAGH,MAAM,aAAa,QAAmC;AACpD,MAAI,OAAO,WAAW,EAAG;EAEzB,MAAM,SAAS,MAAM,KAAK,KAAK,SAAS;AACxC,MAAI;AACF,SAAM,OAAO,MAAM,QAAQ;AAE3B,QAAK,MAAM,KAAK,QAAQ;IAEtB,MAAM,WAAW,EAAE,KAAK,QAAQ,OAAO,GAAG;AAG1C,UAAM,OAAO,MACX;;;;;;;;;;;;;;;gDAgBA;KACE,EAAE;KACF,EAAE;KACF,EAAE;KACF,EAAE;KACF,EAAE;KACF,EAAE;KACF,EAAE;KACF,EAAE;KACF;KACA,EAAE;KACH,CACF;;AAGH,SAAM,OAAO,MAAM,SAAS;WACrB,GAAG;AACV,SAAM,OAAO,MAAM,WAAW;AAC9B,SAAM;YACE;AACR,UAAO,SAAS;;;CAIpB,MAAM,sBAAsB,WAAsC;AAKhE,UAJe,MAAM,KAAK,KAAK,MAC7B,8DACA,CAAC,UAAU,CACZ,EACa,KAAK,KAAK,MAAM,EAAE,KAAK;;CAGvC,MAAM,YAAY,WAAmB,OAAgC;AACnE,MAAI,MAAM,WAAW,EAAG;EACxB,MAAM,SAAS,MAAM,KAAK,KAAK,SAAS;AACxC,MAAI;AACF,SAAM,OAAO,MAAM,QAAQ;AAC3B,QAAK,MAAM,QAAQ,OAAO;AACxB,UAAM,OAAO,MACX,8DACA,CAAC,WAAW,KAAK,CAClB;AACD,UAAM,OAAO,MACX,6DACA,CAAC,WAAW,KAAK,CAClB;;AAEH,SAAM,OAAO,MAAM,SAAS;WACrB,GAAG;AACV,SAAM,OAAO,MAAM,WAAW;AAC9B,SAAM;YACE;AACR,UAAO,SAAS;;;CAIpB,MAAM,sBAAsB,WAAkE;AAC5F,MAAI,cAAc,OAKhB,SAJe,MAAM,KAAK,KAAK,MAC7B,2FACA,CAAC,UAAU,CACZ,EACa;AAKhB,UAHe,MAAM,KAAK,KAAK,MAC7B,sEACD,EACa;;CAGhB,MAAM,gBAAgB,SAAiB,WAAkC;EAGvE,MAAM,SAAS,MADH,eAAe,UAAU,CACZ,KAAK,IAAI,GAAG;AACrC,QAAM,KAAK,KAAK,MACd,8DACA,CAAC,QAAQ,QAAQ,CAClB;;CAOH,MAAM,cAAc,OAAe,MAA+C;EAChF,MAAM,aAAa,MAAM,cAAc;EAGvC,MAAM,UAAU,eAAe,MAAM;AACrC,MAAI,CAAC,QAAS,QAAO,EAAE;EAIvB,MAAM,aAAuB,CAAC,yCAAyC;EACvE,MAAM,SAA8B,CAAC,QAAQ;EAC7C,IAAI,WAAW;AAEf,MAAI,MAAM,cAAc,KAAK,WAAW,SAAS,GAAG;GAClD,MAAM,eAAe,KAAK,WAAW,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AAC3E,cAAW,KAAK,kBAAkB,aAAa,GAAG;AAClD,UAAO,KAAK,GAAG,KAAK,WAAW;;AAGjC,MAAI,MAAM,WAAW,KAAK,QAAQ,SAAS,GAAG;GAC5C,MAAM,eAAe,KAAK,QAAQ,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACxE,cAAW,KAAK,cAAc,aAAa,GAAG;AAC9C,UAAO,KAAK,GAAG,KAAK,QAAQ;;AAG9B,MAAI,MAAM,SAAS,KAAK,MAAM,SAAS,GAAG;GACxC,MAAM,eAAe,KAAK,MAAM,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACtE,cAAW,KAAK,YAAY,aAAa,GAAG;AAC5C,UAAO,KAAK,GAAG,KAAK,MAAM;;AAG5B,SAAO,KAAK,WAAW;EACvB,MAAM,aAAa,IAAI;EAEvB,MAAM,MAAM;;;;;;;;;;;cAWF,WAAW,KAAK,QAAQ,CAAC;;cAEzB,WAAW;;AAGrB,MAAI;AAYF,WAXe,MAAM,KAAK,KAAK,MAS5B,KAAK,OAAO,EAED,KAAK,KAAK,SAAS;IAC/B,WAAW,IAAI;IACf,MAAM,IAAI;IACV,WAAW,IAAI;IACf,SAAS,IAAI;IACb,SAAS,IAAI;IACb,OAAO,IAAI;IACX,MAAM,IAAI;IACV,QAAQ,IAAI;IACb,EAAE;WACI,GAAG;AACV,WAAQ,OAAO,MAAM,uCAAuC,EAAE,IAAI;AAClE,UAAO,EAAE;;;CAQb,MAAM,eAAe,gBAA8B,MAA+C;EAChG,MAAM,aAAa,MAAM,cAAc;EAEvC,MAAM,aAAuB,CAAC,wBAAwB;EACtD,MAAM,SAAuC,EAAE;EAC/C,IAAI,WAAW;EAGf,MAAM,SAAS,MAAM,MAAM,KAAK,eAAe,CAAC,KAAK,IAAI,GAAG;AAC5D,SAAO,KAAK,OAAO;EACnB,MAAM,WAAW,IAAI;AAErB,MAAI,MAAM,cAAc,KAAK,WAAW,SAAS,GAAG;GAClD,MAAM,eAAe,KAAK,WAAW,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AAC3E,cAAW,KAAK,kBAAkB,aAAa,GAAG;AAClD,UAAO,KAAK,GAAG,KAAK,WAAW;;AAGjC,MAAI,MAAM,WAAW,KAAK,QAAQ,SAAS,GAAG;GAC5C,MAAM,eAAe,KAAK,QAAQ,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACxE,cAAW,KAAK,cAAc,aAAa,GAAG;AAC9C,UAAO,KAAK,GAAG,KAAK,QAAQ;;AAG9B,MAAI,MAAM,SAAS,KAAK,MAAM,SAAS,GAAG;GACxC,MAAM,eAAe,KAAK,MAAM,UAAU,IAAI,aAAa,CAAC,KAAK,KAAK;AACtE,cAAW,KAAK,YAAY,aAAa,GAAG;AAC5C,UAAO,KAAK,GAAG,KAAK,MAAM;;AAG5B,SAAO,KAAK,WAAW;EACvB,MAAM,aAAa,IAAI;EAGvB,MAAM,MAAM;;;;;;;;;6BASa,SAAS;;cAExB,WAAW,KAAK,QAAQ,CAAC;+BACR,SAAS;cAC1B,WAAW;;AAGrB,MAAI;GACF,MAAM,SAAS,MAAM,KAAK,KAAK,MAS5B,KAAK,OAAO;GAEf,MAAM,WAAW,MAAM,YAAY;AAEnC,UAAO,OAAO,KACX,KAAK,SAAS;IACb,WAAW,IAAI;IACf,MAAM,IAAI;IACV,WAAW,IAAI;IACf,SAAS,IAAI;IACb,SAAS,IAAI;IACb,OAAO,IAAI;IACX,MAAM,IAAI;IACV,QAAQ,IAAI;IACb,EAAE,CACF,QAAQ,MAAM,EAAE,SAAS,SAAS;WAC9B,GAAG;AACV,WAAQ,OAAO,MAAM,wCAAwC,EAAE,IAAI;AACnE,UAAO,EAAE;;;;;;;AAYf,SAAS,eAAe,KAAuB;CAC7C,MAAM,SAAmB,EAAE;AAC3B,MAAK,IAAI,IAAI,GAAG,IAAI,IAAI,QAAQ,KAAK,EACnC,QAAO,KAAK,IAAI,YAAY,EAAE,CAAC;AAEjC,QAAO;;;;;;;;;;;;;;AAeT,SAAS,eAAe,OAAuB;CAC7C,MAAM,aAAa,IAAI,IAAI;EACzB;EAAK;EAAM;EAAO;EAAO;EAAM;EAAM;EAAM;EAAQ;EAAO;EAC1D;EAAM;EAAO;EAAQ;EAAO;EAAQ;EAAM;EAAO;EAAO;EACxD;EAAO;EAAK;EAAM;EAAM;EAAM;EAAM;EAAO;EAAM;EAAM;EACvD;EAAM;EAAM;EAAM;EAAO;EAAO;EAAO;EAAM;EAAQ;EACrD;EAAS;EAAQ;EAAQ;EAAQ;EAAM;EAAM;EAAM;EAAO;EAC1D;EAAQ;EAAQ;EAAQ;EAAO;EAAQ;EAAQ;EAAO;EACvD,CAAC;CAEF,MAAM,SAAS,MACZ,aAAa,CACb,MAAM,cAAc,CACpB,OAAO,QAAQ,CACf,QAAQ,MAAM,EAAE,UAAU,EAAE,CAC5B,QAAQ,MAAM,CAAC,WAAW,IAAI,EAAE,CAAC,CAEjC,KAAK,MAAM,EAAE,QAAQ,MAAM,KAAK,CAAC,QAAQ,aAAa,GAAG,CAAC,CAC1D,OAAO,QAAQ;AAElB,KAAI,OAAO,WAAW,EAGpB,QADY,MAAM,QAAQ,eAAe,IAAI,CAAC,MAAM,CAAC,MAAM,MAAM,CAAC,OAAO,QAAQ,CAAC,KAAK,MAAM,IAC/E;AAMhB,QAAO,OAAO,KAAK,MAAM"}

package/dist/{sqlite-CymLKiDE.mjs → sqlite-WWBq7_2C.mjs}

@@ -64,6 +64,23 @@ var SQLiteBackend = class {
 			}
 		})();
 	}
+	async getDistinctChunkPaths(projectId) {
+		return this.db.prepare("SELECT DISTINCT path FROM memory_chunks WHERE project_id = ?").all(projectId).map((r) => r.path);
+	}
+	async deletePaths(projectId, paths) {
+		if (paths.length === 0) return;
+		const deleteFts = this.db.prepare("DELETE FROM memory_fts WHERE id = ?");
+		const deleteChunks = this.db.prepare("DELETE FROM memory_chunks WHERE project_id = ? AND path = ?");
+		const deleteFile = this.db.prepare("DELETE FROM memory_files WHERE project_id = ? AND path = ?");
+		this.db.transaction(() => {
+			for (const path of paths) {
+				const ids = this.db.prepare("SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?").all(projectId, path);
+				for (const { id } of ids) deleteFts.run(id);
+				deleteChunks.run(projectId, path);
+				deleteFile.run(projectId, path);
+			}
+		})();
+	}
 	async getUnembeddedChunkIds(projectId) {
 		const conditions = ["embedding IS NULL"];
 		const params = [];
@@ -87,4 +104,4 @@ var SQLiteBackend = class {
 
 //#endregion
 export { SQLiteBackend };
-//# sourceMappingURL=sqlite-CymLKiDE.mjs.map
+//# sourceMappingURL=sqlite-WWBq7_2C.mjs.map

package/dist/{sqlite-CymLKiDE.mjs.map → sqlite-WWBq7_2C.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"sqlite-CymLKiDE.mjs","names":[],"sources":["../src/storage/sqlite.ts"],"sourcesContent":["/**\n * SQLiteBackend — wraps the existing better-sqlite3 federation.db\n * behind the StorageBackend interface.\n *\n * This is a thin adapter.  The heavy lifting is all in the existing\n * memory/indexer.ts and memory/search.ts code; we just provide a\n * backend-agnostic surface so the daemon and tools can call either\n * SQLite or Postgres transparently.\n */\n\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow, FileRow, FederationStats } from \"./interface.js\";\nimport type { SearchResult, SearchOptions } from \"../memory/search.js\";\nimport { searchMemory, searchMemorySemantic } from \"../memory/search.js\";\n\nexport class SQLiteBackend implements StorageBackend {\n  readonly backendType = \"sqlite\" as const;\n\n  private db: Database;\n\n  constructor(db: Database) {\n    this.db = db;\n  }\n\n  /**\n   * Expose the raw better-sqlite3 Database handle.\n   * Used by the daemon to pass to indexAll() which still uses the synchronous API directly.\n   */\n  getRawDb(): Database {\n    return this.db;\n  }\n\n  // -------------------------------------------------------------------------\n  // Lifecycle\n  // -------------------------------------------------------------------------\n\n  async close(): Promise<void> {\n    try {\n      this.db.close();\n    } catch {\n      // ignore\n    }\n  }\n\n  async getStats(): Promise<FederationStats> {\n    const files = (\n      this.db.prepare(\"SELECT COUNT(*) AS n FROM memory_files\").get() as { n: number }\n    ).n;\n    const chunks = (\n      this.db.prepare(\"SELECT COUNT(*) AS n FROM memory_chunks\").get() as { n: number }\n    ).n;\n    return { files, chunks };\n  }\n\n  // -------------------------------------------------------------------------\n  // File tracking\n  // -------------------------------------------------------------------------\n\n  async getFileHash(projectId: number, path: string): Promise<string | undefined> {\n    const row = this.db\n      .prepare(\"SELECT hash FROM memory_files WHERE project_id = ? AND path = ?\")\n      .get(projectId, path) as { hash: string } | undefined;\n    return row?.hash;\n  }\n\n  async upsertFile(file: FileRow): Promise<void> {\n    this.db\n      .prepare(\n        `INSERT INTO memory_files (project_id, path, source, tier, hash, mtime, size)\n         VALUES (?, ?, ?, ?, ?, ?, ?)\n         ON CONFLICT(project_id, path) DO UPDATE SET\n           source = excluded.source,\n           tier   = excluded.tier,\n           hash   = excluded.hash,\n           mtime  = excluded.mtime,\n           size   = excluded.size`\n      )\n      .run(file.projectId, file.path, file.source, file.tier, file.hash, file.mtime, file.size);\n  }\n\n  // -------------------------------------------------------------------------\n  // Chunk management\n  // -------------------------------------------------------------------------\n\n  async getChunkIds(projectId: number, path: string): Promise<string[]> {\n    const rows = this.db\n      .prepare(\"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\")\n      .all(projectId, path) as Array<{ id: string }>;\n    return rows.map((r) => r.id);\n  }\n\n  async deleteChunksForFile(projectId: number, path: string): Promise<void> {\n    const ids = await this.getChunkIds(projectId, path);\n    const deleteFts = this.db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n    const deleteChunks = this.db.prepare(\n      \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\"\n    );\n    this.db.transaction(() => {\n      for (const id of ids) {\n        deleteFts.run(id);\n      }\n      deleteChunks.run(projectId, path);\n    })();\n  }\n\n  async insertChunks(chunks: ChunkRow[]): Promise<void> {\n    if (chunks.length === 0) return;\n\n    const insertChunk = this.db.prepare(\n      `INSERT INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`\n    );\n    const insertFts = this.db.prepare(\n      `INSERT INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n       VALUES (?, ?, ?, ?, ?, ?, ?, ?)`\n    );\n\n    this.db.transaction(() => {\n      for (const c of chunks) {\n        insertChunk.run(\n          c.id,\n          c.projectId,\n          c.source,\n          c.tier,\n          c.path,\n          c.startLine,\n          c.endLine,\n          c.hash,\n          c.text,\n          c.updatedAt\n        );\n        insertFts.run(\n          c.text,\n          c.id,\n          c.projectId,\n          c.path,\n          c.source,\n          c.tier,\n          c.startLine,\n          c.endLine\n        );\n      }\n    })();\n  }\n\n  async getUnembeddedChunkIds(projectId?: number): Promise<Array<{ id: string; text: string }>> {\n    const conditions = [\"embedding IS NULL\"];\n    const params: (string | number)[] = [];\n\n    if (projectId !== undefined) {\n      conditions.push(\"project_id = ?\");\n      params.push(projectId);\n    }\n\n    const where = \"WHERE \" + conditions.join(\" AND \");\n    const rows = this.db\n      .prepare(`SELECT id, text FROM memory_chunks ${where} ORDER BY id`)\n      .all(...params) as Array<{ id: string; text: string }>;\n    return rows;\n  }\n\n  async updateEmbedding(chunkId: string, embedding: Buffer): Promise<void> {\n    this.db\n      .prepare(\"UPDATE memory_chunks SET embedding = ? WHERE id = ?\")\n      .run(embedding, chunkId);\n  }\n\n  // -------------------------------------------------------------------------\n  // Search\n  // -------------------------------------------------------------------------\n\n  async searchKeyword(query: string, opts?: SearchOptions): Promise<SearchResult[]> {\n    return searchMemory(this.db, query, opts);\n  }\n\n  async searchSemantic(queryEmbedding: Float32Array, opts?: SearchOptions): Promise<SearchResult[]> {\n    return searchMemorySemantic(this.db, queryEmbedding, opts);\n  }\n}\n"],"mappings":";;;;AAeA,IAAa,gBAAb,MAAqD;CACnD,AAAS,cAAc;CAEvB,AAAQ;CAER,YAAY,IAAc;AACxB,OAAK,KAAK;;;;;;CAOZ,WAAqB;AACnB,SAAO,KAAK;;CAOd,MAAM,QAAuB;AAC3B,MAAI;AACF,QAAK,GAAG,OAAO;UACT;;CAKV,MAAM,WAAqC;AAOzC,SAAO;GAAE,OALP,KAAK,GAAG,QAAQ,yCAAyC,CAAC,KAAK,CAC/D;GAIc,QAFd,KAAK,GAAG,QAAQ,0CAA0C,CAAC,KAAK,CAChE;GACsB;;CAO1B,MAAM,YAAY,WAAmB,MAA2C;AAI9E,SAHY,KAAK,GACd,QAAQ,kEAAkE,CAC1E,IAAI,WAAW,KAAK,EACX;;CAGd,MAAM,WAAW,MAA8B;AAC7C,OAAK,GACF,QACC;;;;;;;mCAQD,CACA,IAAI,KAAK,WAAW,KAAK,MAAM,KAAK,QAAQ,KAAK,MAAM,KAAK,MAAM,KAAK,OAAO,KAAK,KAAK;;CAO7F,MAAM,YAAY,WAAmB,MAAiC;AAIpE,SAHa,KAAK,GACf,QAAQ,iEAAiE,CACzE,IAAI,WAAW,KAAK,CACX,KAAK,MAAM,EAAE,GAAG;;CAG9B,MAAM,oBAAoB,WAAmB,MAA6B;EACxE,MAAM,MAAM,MAAM,KAAK,YAAY,WAAW,KAAK;EACnD,MAAM,YAAY,KAAK,GAAG,QAAQ,sCAAsC;EACxE,MAAM,eAAe,KAAK,GAAG,QAC3B,8DACD;AACD,OAAK,GAAG,kBAAkB;AACxB,QAAK,MAAM,MAAM,IACf,WAAU,IAAI,GAAG;AAEnB,gBAAa,IAAI,WAAW,KAAK;IACjC,EAAE;;CAGN,MAAM,aAAa,QAAmC;AACpD,MAAI,OAAO,WAAW,EAAG;EAEzB,MAAM,cAAc,KAAK,GAAG,QAC1B;8CAED;EACD,MAAM,YAAY,KAAK,GAAG,QACxB;wCAED;AAED,OAAK,GAAG,kBAAkB;AACxB,QAAK,MAAM,KAAK,QAAQ;AACtB,gBAAY,IACV,EAAE,IACF,EAAE,WACF,EAAE,QACF,EAAE,MACF,EAAE,MACF,EAAE,WACF,EAAE,SACF,EAAE,MACF,EAAE,MACF,EAAE,UACH;AACD,cAAU,IACR,EAAE,MACF,EAAE,IACF,EAAE,WACF,EAAE,MACF,EAAE,QACF,EAAE,MACF,EAAE,WACF,EAAE,QACH;;IAEH,EAAE;;CAGN,MAAM,sBAAsB,WAAkE;EAC5F,MAAM,aAAa,CAAC,oBAAoB;EACxC,MAAM,SAA8B,EAAE;AAEtC,MAAI,cAAc,QAAW;AAC3B,cAAW,KAAK,iBAAiB;AACjC,UAAO,KAAK,UAAU;;EAGxB,MAAM,QAAQ,WAAW,WAAW,KAAK,QAAQ;AAIjD,SAHa,KAAK,GACf,QAAQ,sCAAsC,MAAM,cAAc,CAClE,IAAI,GAAG,OAAO;;CAInB,MAAM,gBAAgB,SAAiB,WAAkC;AACvE,OAAK,GACF,QAAQ,sDAAsD,CAC9D,IAAI,WAAW,QAAQ;;CAO5B,MAAM,cAAc,OAAe,MAA+C;AAChF,SAAO,aAAa,KAAK,IAAI,OAAO,KAAK;;CAG3C,MAAM,eAAe,gBAA8B,MAA+C;AAChG,SAAO,qBAAqB,KAAK,IAAI,gBAAgB,KAAK"}
+{"version":3,"file":"sqlite-WWBq7_2C.mjs","names":[],"sources":["../src/storage/sqlite.ts"],"sourcesContent":["/**\n * SQLiteBackend — wraps the existing better-sqlite3 federation.db\n * behind the StorageBackend interface.\n *\n * This is a thin adapter.  The heavy lifting is all in the existing\n * memory/indexer.ts and memory/search.ts code; we just provide a\n * backend-agnostic surface so the daemon and tools can call either\n * SQLite or Postgres transparently.\n */\n\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow, FileRow, FederationStats } from \"./interface.js\";\nimport type { SearchResult, SearchOptions } from \"../memory/search.js\";\nimport { searchMemory, searchMemorySemantic } from \"../memory/search.js\";\n\nexport class SQLiteBackend implements StorageBackend {\n  readonly backendType = \"sqlite\" as const;\n\n  private db: Database;\n\n  constructor(db: Database) {\n    this.db = db;\n  }\n\n  /**\n   * Expose the raw better-sqlite3 Database handle.\n   * Used by the daemon to pass to indexAll() which still uses the synchronous API directly.\n   */\n  getRawDb(): Database {\n    return this.db;\n  }\n\n  // -------------------------------------------------------------------------\n  // Lifecycle\n  // -------------------------------------------------------------------------\n\n  async close(): Promise<void> {\n    try {\n      this.db.close();\n    } catch {\n      // ignore\n    }\n  }\n\n  async getStats(): Promise<FederationStats> {\n    const files = (\n      this.db.prepare(\"SELECT COUNT(*) AS n FROM memory_files\").get() as { n: number }\n    ).n;\n    const chunks = (\n      this.db.prepare(\"SELECT COUNT(*) AS n FROM memory_chunks\").get() as { n: number }\n    ).n;\n    return { files, chunks };\n  }\n\n  // -------------------------------------------------------------------------\n  // File tracking\n  // -------------------------------------------------------------------------\n\n  async getFileHash(projectId: number, path: string): Promise<string | undefined> {\n    const row = this.db\n      .prepare(\"SELECT hash FROM memory_files WHERE project_id = ? AND path = ?\")\n      .get(projectId, path) as { hash: string } | undefined;\n    return row?.hash;\n  }\n\n  async upsertFile(file: FileRow): Promise<void> {\n    this.db\n      .prepare(\n        `INSERT INTO memory_files (project_id, path, source, tier, hash, mtime, size)\n         VALUES (?, ?, ?, ?, ?, ?, ?)\n         ON CONFLICT(project_id, path) DO UPDATE SET\n           source = excluded.source,\n           tier   = excluded.tier,\n           hash   = excluded.hash,\n           mtime  = excluded.mtime,\n           size   = excluded.size`\n      )\n      .run(file.projectId, file.path, file.source, file.tier, file.hash, file.mtime, file.size);\n  }\n\n  // -------------------------------------------------------------------------\n  // Chunk management\n  // -------------------------------------------------------------------------\n\n  async getChunkIds(projectId: number, path: string): Promise<string[]> {\n    const rows = this.db\n      .prepare(\"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\")\n      .all(projectId, path) as Array<{ id: string }>;\n    return rows.map((r) => r.id);\n  }\n\n  async deleteChunksForFile(projectId: number, path: string): Promise<void> {\n    const ids = await this.getChunkIds(projectId, path);\n    const deleteFts = this.db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n    const deleteChunks = this.db.prepare(\n      \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\"\n    );\n    this.db.transaction(() => {\n      for (const id of ids) {\n        deleteFts.run(id);\n      }\n      deleteChunks.run(projectId, path);\n    })();\n  }\n\n  async insertChunks(chunks: ChunkRow[]): Promise<void> {\n    if (chunks.length === 0) return;\n\n    const insertChunk = this.db.prepare(\n      `INSERT INTO memory_chunks (id, project_id, source, tier, path, start_line, end_line, hash, text, updated_at)\n       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`\n    );\n    const insertFts = this.db.prepare(\n      `INSERT INTO memory_fts (text, id, project_id, path, source, tier, start_line, end_line)\n       VALUES (?, ?, ?, ?, ?, ?, ?, ?)`\n    );\n\n    this.db.transaction(() => {\n      for (const c of chunks) {\n        insertChunk.run(\n          c.id,\n          c.projectId,\n          c.source,\n          c.tier,\n          c.path,\n          c.startLine,\n          c.endLine,\n          c.hash,\n          c.text,\n          c.updatedAt\n        );\n        insertFts.run(\n          c.text,\n          c.id,\n          c.projectId,\n          c.path,\n          c.source,\n          c.tier,\n          c.startLine,\n          c.endLine\n        );\n      }\n    })();\n  }\n\n  async getDistinctChunkPaths(projectId: number): Promise<string[]> {\n    const rows = this.db\n      .prepare(\"SELECT DISTINCT path FROM memory_chunks WHERE project_id = ?\")\n      .all(projectId) as Array<{ path: string }>;\n    return rows.map((r) => r.path);\n  }\n\n  async deletePaths(projectId: number, paths: string[]): Promise<void> {\n    if (paths.length === 0) return;\n    const deleteFts = this.db.prepare(\"DELETE FROM memory_fts WHERE id = ?\");\n    const deleteChunks = this.db.prepare(\n      \"DELETE FROM memory_chunks WHERE project_id = ? AND path = ?\"\n    );\n    const deleteFile = this.db.prepare(\n      \"DELETE FROM memory_files WHERE project_id = ? AND path = ?\"\n    );\n    this.db.transaction(() => {\n      for (const path of paths) {\n        const ids = this.db\n          .prepare(\"SELECT id FROM memory_chunks WHERE project_id = ? AND path = ?\")\n          .all(projectId, path) as Array<{ id: string }>;\n        for (const { id } of ids) {\n          deleteFts.run(id);\n        }\n        deleteChunks.run(projectId, path);\n        deleteFile.run(projectId, path);\n      }\n    })();\n  }\n\n  async getUnembeddedChunkIds(projectId?: number): Promise<Array<{ id: string; text: string }>> {\n    const conditions = [\"embedding IS NULL\"];\n    const params: (string | number)[] = [];\n\n    if (projectId !== undefined) {\n      conditions.push(\"project_id = ?\");\n      params.push(projectId);\n    }\n\n    const where = \"WHERE \" + conditions.join(\" AND \");\n    const rows = this.db\n      .prepare(`SELECT id, text FROM memory_chunks ${where} ORDER BY id`)\n      .all(...params) as Array<{ id: string; text: string }>;\n    return rows;\n  }\n\n  async updateEmbedding(chunkId: string, embedding: Buffer): Promise<void> {\n    this.db\n      .prepare(\"UPDATE memory_chunks SET embedding = ? WHERE id = ?\")\n      .run(embedding, chunkId);\n  }\n\n  // -------------------------------------------------------------------------\n  // Search\n  // -------------------------------------------------------------------------\n\n  async searchKeyword(query: string, opts?: SearchOptions): Promise<SearchResult[]> {\n    return searchMemory(this.db, query, opts);\n  }\n\n  async searchSemantic(queryEmbedding: Float32Array, opts?: SearchOptions): Promise<SearchResult[]> {\n    return searchMemorySemantic(this.db, queryEmbedding, opts);\n  }\n}\n"],"mappings":";;;;AAeA,IAAa,gBAAb,MAAqD;CACnD,AAAS,cAAc;CAEvB,AAAQ;CAER,YAAY,IAAc;AACxB,OAAK,KAAK;;;;;;CAOZ,WAAqB;AACnB,SAAO,KAAK;;CAOd,MAAM,QAAuB;AAC3B,MAAI;AACF,QAAK,GAAG,OAAO;UACT;;CAKV,MAAM,WAAqC;AAOzC,SAAO;GAAE,OALP,KAAK,GAAG,QAAQ,yCAAyC,CAAC,KAAK,CAC/D;GAIc,QAFd,KAAK,GAAG,QAAQ,0CAA0C,CAAC,KAAK,CAChE;GACsB;;CAO1B,MAAM,YAAY,WAAmB,MAA2C;AAI9E,SAHY,KAAK,GACd,QAAQ,kEAAkE,CAC1E,IAAI,WAAW,KAAK,EACX;;CAGd,MAAM,WAAW,MAA8B;AAC7C,OAAK,GACF,QACC;;;;;;;mCAQD,CACA,IAAI,KAAK,WAAW,KAAK,MAAM,KAAK,QAAQ,KAAK,MAAM,KAAK,MAAM,KAAK,OAAO,KAAK,KAAK;;CAO7F,MAAM,YAAY,WAAmB,MAAiC;AAIpE,SAHa,KAAK,GACf,QAAQ,iEAAiE,CACzE,IAAI,WAAW,KAAK,CACX,KAAK,MAAM,EAAE,GAAG;;CAG9B,MAAM,oBAAoB,WAAmB,MAA6B;EACxE,MAAM,MAAM,MAAM,KAAK,YAAY,WAAW,KAAK;EACnD,MAAM,YAAY,KAAK,GAAG,QAAQ,sCAAsC;EACxE,MAAM,eAAe,KAAK,GAAG,QAC3B,8DACD;AACD,OAAK,GAAG,kBAAkB;AACxB,QAAK,MAAM,MAAM,IACf,WAAU,IAAI,GAAG;AAEnB,gBAAa,IAAI,WAAW,KAAK;IACjC,EAAE;;CAGN,MAAM,aAAa,QAAmC;AACpD,MAAI,OAAO,WAAW,EAAG;EAEzB,MAAM,cAAc,KAAK,GAAG,QAC1B;8CAED;EACD,MAAM,YAAY,KAAK,GAAG,QACxB;wCAED;AAED,OAAK,GAAG,kBAAkB;AACxB,QAAK,MAAM,KAAK,QAAQ;AACtB,gBAAY,IACV,EAAE,IACF,EAAE,WACF,EAAE,QACF,EAAE,MACF,EAAE,MACF,EAAE,WACF,EAAE,SACF,EAAE,MACF,EAAE,MACF,EAAE,UACH;AACD,cAAU,IACR,EAAE,MACF,EAAE,IACF,EAAE,WACF,EAAE,MACF,EAAE,QACF,EAAE,MACF,EAAE,WACF,EAAE,QACH;;IAEH,EAAE;;CAGN,MAAM,sBAAsB,WAAsC;AAIhE,SAHa,KAAK,GACf,QAAQ,+DAA+D,CACvE,IAAI,UAAU,CACL,KAAK,MAAM,EAAE,KAAK;;CAGhC,MAAM,YAAY,WAAmB,OAAgC;AACnE,MAAI,MAAM,WAAW,EAAG;EACxB,MAAM,YAAY,KAAK,GAAG,QAAQ,sCAAsC;EACxE,MAAM,eAAe,KAAK,GAAG,QAC3B,8DACD;EACD,MAAM,aAAa,KAAK,GAAG,QACzB,6DACD;AACD,OAAK,GAAG,kBAAkB;AACxB,QAAK,MAAM,QAAQ,OAAO;IACxB,MAAM,MAAM,KAAK,GACd,QAAQ,iEAAiE,CACzE,IAAI,WAAW,KAAK;AACvB,SAAK,MAAM,EAAE,QAAQ,IACnB,WAAU,IAAI,GAAG;AAEnB,iBAAa,IAAI,WAAW,KAAK;AACjC,eAAW,IAAI,WAAW,KAAK;;IAEjC,EAAE;;CAGN,MAAM,sBAAsB,WAAkE;EAC5F,MAAM,aAAa,CAAC,oBAAoB;EACxC,MAAM,SAA8B,EAAE;AAEtC,MAAI,cAAc,QAAW;AAC3B,cAAW,KAAK,iBAAiB;AACjC,UAAO,KAAK,UAAU;;EAGxB,MAAM,QAAQ,WAAW,WAAW,KAAK,QAAQ;AAIjD,SAHa,KAAK,GACf,QAAQ,sCAAsC,MAAM,cAAc,CAClE,IAAI,GAAG,OAAO;;CAInB,MAAM,gBAAgB,SAAiB,WAAkC;AACvE,OAAK,GACF,QAAQ,sDAAsD,CAC9D,IAAI,WAAW,QAAQ;;CAO5B,MAAM,cAAc,OAAe,MAA+C;AAChF,SAAO,aAAa,KAAK,IAAI,OAAO,KAAK;;CAG3C,MAAM,eAAe,gBAA8B,MAA+C;AAChG,SAAO,qBAAqB,KAAK,IAAI,gBAAgB,KAAK"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tekmidian/pai",
-  "version": "0.5.1",
+  "version": "0.5.2",
   "description": "PAI Knowledge OS — Personal AI Infrastructure with federated memory and project management",
   "type": "module",
   "main": "dist/index.mjs",

package/dist/indexer-backend-DQO-FqAI.mjs.map

@@ -1 +0,0 @@
-{"version":3,"file":"indexer-backend-DQO-FqAI.mjs","names":[],"sources":["../src/memory/indexer-backend.ts"],"sourcesContent":["/**\n * Backend-aware indexer for PAI federation memory.\n *\n * This module provides the same functionality as indexer.ts but writes\n * through the 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 indexer.ts directly (which is faster for SQLite\n * due to synchronous transactions).\n */\n\nimport { createHash } from \"node:crypto\";\nimport { readFileSync, statSync, readdirSync, existsSync } from \"node:fs\";\nimport { join, relative, basename, normalize } from \"node:path\";\n\n// ---------------------------------------------------------------------------\n// Session title parsing\n// ---------------------------------------------------------------------------\n\nconst SESSION_TITLE_RE = /^(\\d{4})\\s*-\\s*(\\d{4}-\\d{2}-\\d{2})\\s*-\\s*(.+)\\.md$/;\n\n/**\n * Parse a session title from a Notes filename.\n * Format: \"NNNN - YYYY-MM-DD - Descriptive Title.md\"\n * Returns a synthetic chunk text like \"Session #0086 2026-02-23: Pai Daemon Background Service\"\n * or null if the filename doesn't match the expected pattern.\n */\nexport function parseSessionTitleChunk(fileName: string): string | null {\n  const m = SESSION_TITLE_RE.exec(fileName);\n  if (!m) return null;\n  const [, num, date, title] = m;\n  return `Session #${num} ${date}: ${title}`;\n}\nimport { homedir } from \"node:os\";\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow } from \"../storage/interface.js\";\nimport type { IndexResult } from \"./indexer.js\";\nimport { chunkMarkdown } from \"./chunker.js\";\nimport { detectTier } from \"./indexer.js\";\n\n// ---------------------------------------------------------------------------\n// Constants (mirrored from indexer.ts)\n// ---------------------------------------------------------------------------\n\nconst MAX_FILES_PER_PROJECT = 5_000;\nconst MAX_WALK_DEPTH = 6;\nconst INDEX_YIELD_EVERY = 10;\n\n/**\n * Directories to ALWAYS skip, at any depth, during any directory walk.\n * These are build artifacts, dependency trees, and VCS internals that\n * should never be indexed regardless of where they appear in the tree.\n */\nconst ALWAYS_SKIP_DIRS = new Set([\n  // Version control\n  \".git\",\n  // Dependency directories (any language)\n  \"node_modules\",\n  \"vendor\",\n  \"Pods\",              // CocoaPods (iOS/macOS)\n  // Build / compile output\n  \"dist\",\n  \"build\",\n  \"out\",\n  \"DerivedData\",       // Xcode\n  \".next\",             // Next.js\n  // Python virtual environments and caches\n  \".venv\",\n  \"venv\",\n  \"__pycache__\",\n  // General caches\n  \".cache\",\n  \".bun\",\n]);\n\nconst ROOT_SCAN_SKIP_DIRS = new Set([\n  \"memory\", \"Notes\", \".claude\", \".DS_Store\",\n  ...ALWAYS_SKIP_DIRS,\n]);\n\nconst CONTENT_SCAN_SKIP_DIRS = new Set([\n  \"Library\", \"Applications\", \"Music\", \"Movies\", \"Pictures\", \"Desktop\",\n  \"Downloads\", \"Public\", \"coverage\",\n  ...ALWAYS_SKIP_DIRS,\n]);\n\n// ---------------------------------------------------------------------------\n// Helpers (same logic as indexer.ts)\n// ---------------------------------------------------------------------------\n\nfunction sha256File(content: string): string {\n  return createHash(\"sha256\").update(content).digest(\"hex\");\n}\n\nfunction chunkId(\n  projectId: number,\n  path: string,\n  chunkIndex: number,\n  startLine: number,\n  endLine: number,\n): string {\n  return createHash(\"sha256\")\n    .update(`${projectId}:${path}:${chunkIndex}:${startLine}:${endLine}`)\n    .digest(\"hex\");\n}\n\nfunction walkMdFiles(\n  dir: string,\n  acc?: string[],\n  cap = MAX_FILES_PER_PROJECT,\n  depth = 0,\n): string[] {\n  const results = acc ?? [];\n  if (!existsSync(dir)) return results;\n  if (results.length >= cap) return results;\n  if (depth > MAX_WALK_DEPTH) return results;\n  try {\n    for (const entry of readdirSync(dir, { withFileTypes: true })) {\n      if (results.length >= cap) break;\n      if (entry.isSymbolicLink()) continue;\n      // Skip known junk directories at every recursion depth\n      if (ALWAYS_SKIP_DIRS.has(entry.name)) continue;\n      const full = join(dir, entry.name);\n      if (entry.isDirectory()) {\n        walkMdFiles(full, results, cap, depth + 1);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        results.push(full);\n      }\n    }\n  } catch { /* skip unreadable */ }\n  return results;\n}\n\nfunction walkContentFiles(rootPath: string): string[] {\n  if (!existsSync(rootPath)) return [];\n  const results: string[] = [];\n  try {\n    for (const entry of readdirSync(rootPath, { withFileTypes: true })) {\n      if (results.length >= MAX_FILES_PER_PROJECT) break;\n      if (entry.isSymbolicLink()) continue;\n      if (ROOT_SCAN_SKIP_DIRS.has(entry.name)) continue;\n      if (CONTENT_SCAN_SKIP_DIRS.has(entry.name)) continue;\n      const full = join(rootPath, entry.name);\n      if (entry.isDirectory()) {\n        walkMdFiles(full, results, MAX_FILES_PER_PROJECT);\n      } else if (entry.isFile() && entry.name.endsWith(\".md\")) {\n        if (entry.name !== \"MEMORY.md\") results.push(full);\n      }\n    }\n  } catch { /* skip */ }\n  return results;\n}\n\nfunction isPathTooBroadForContentScan(rootPath: string): boolean {\n  const normalized = normalize(rootPath);\n  const home = homedir();\n  if (home.startsWith(normalized) || normalized === \"/\") return true;\n  if (normalized.startsWith(home)) {\n    const rel = normalized.slice(home.length).replace(/^\\//, \"\");\n    const depth = rel ? rel.split(\"/\").length : 0;\n    if (depth === 0) return true;\n  }\n  if (existsSync(join(normalized, \".git\"))) return true;\n  return false;\n}\n\nfunction yieldToEventLoop(): Promise<void> {\n  return new Promise((resolve) => setImmediate(resolve));\n}\n\n// ---------------------------------------------------------------------------\n// 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: parse titles from Notes filenames and insert\n  // as high-signal chunks so session names are searchable via BM25 and embeddings.\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 = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      const titleChunk: import(\"../storage/interface.js\").ChunkRow = {\n        id, projectId, source: \"notes\", tier: \"session\",\n        path: syntheticPath, startLine: 0, endLine: 0,\n        hash, text, updatedAt, embedding: null,\n      };\n      await backend.insertChunks([titleChunk]);\n    }\n  }\n\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        const titleChunk: import(\"../storage/interface.js\").ChunkRow = {\n          id, projectId, source: \"notes\", tier: \"session\",\n          path: syntheticPath, startLine: 0, endLine: 0,\n          hash, text, updatedAt, embedding: null,\n        };\n        await backend.insertChunks([titleChunk]);\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    const changed = await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier);\n\n    if (changed) {\n      // Count chunks — we know we just inserted them, count from the chunk IDs\n      const ids = await backend.getChunkIds(projectId, relPath);\n      result.filesProcessed++;\n      result.chunksCreated += ids.length;\n    } else {\n      result.filesSkipped++;\n    }\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation via StorageBackend\n// ---------------------------------------------------------------------------\n\nconst EMBED_BATCH_SIZE = 50;\nconst EMBED_YIELD_EVERY = 10;\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): 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  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 } = batch[j];\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    }\n\n    embedded += batch.length;\n    process.stderr.write(\n      `[pai-daemon] Embedded ${embedded}/${total} chunks\\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":";;;;;;;;;;;;;;;;;;AAmBA,MAAM,mBAAmB;;;;;;;AAQzB,SAAgB,uBAAuB,UAAiC;CACtE,MAAM,IAAI,iBAAiB,KAAK,SAAS;AACzC,KAAI,CAAC,EAAG,QAAO;CACf,MAAM,GAAG,KAAK,MAAM,SAAS;AAC7B,QAAO,YAAY,IAAI,GAAG,KAAK,IAAI;;AAarC,MAAM,wBAAwB;AAC9B,MAAM,iBAAiB;AACvB,MAAM,oBAAoB;;;;;;AAO1B,MAAM,mBAAmB,IAAI,IAAI;CAE/B;CAEA;CACA;CACA;CAEA;CACA;CACA;CACA;CACA;CAEA;CACA;CACA;CAEA;CACA;CACD,CAAC;AAEF,MAAM,sBAAsB,IAAI,IAAI;CAClC;CAAU;CAAS;CAAW;CAC9B,GAAG;CACJ,CAAC;AAEF,MAAM,yBAAyB,IAAI,IAAI;CACrC;CAAW;CAAgB;CAAS;CAAU;CAAY;CAC1D;CAAa;CAAU;CACvB,GAAG;CACJ,CAAC;AAMF,SAAS,WAAW,SAAyB;AAC3C,QAAO,WAAW,SAAS,CAAC,OAAO,QAAQ,CAAC,OAAO,MAAM;;AAG3D,SAAS,QACP,WACA,MACA,YACA,WACA,SACQ;AACR,QAAO,WAAW,SAAS,CACxB,OAAO,GAAG,UAAU,GAAG,KAAK,GAAG,WAAW,GAAG,UAAU,GAAG,UAAU,CACpE,OAAO,MAAM;;AAGlB,SAAS,YACP,KACA,KACA,MAAM,uBACN,QAAQ,GACE;CACV,MAAM,UAAU,OAAO,EAAE;AACzB,KAAI,CAAC,WAAW,IAAI,CAAE,QAAO;AAC7B,KAAI,QAAQ,UAAU,IAAK,QAAO;AAClC,KAAI,QAAQ,eAAgB,QAAO;AACnC,KAAI;AACF,OAAK,MAAM,SAAS,YAAY,KAAK,EAAE,eAAe,MAAM,CAAC,EAAE;AAC7D,OAAI,QAAQ,UAAU,IAAK;AAC3B,OAAI,MAAM,gBAAgB,CAAE;AAE5B,OAAI,iBAAiB,IAAI,MAAM,KAAK,CAAE;GACtC,MAAM,OAAO,KAAK,KAAK,MAAM,KAAK;AAClC,OAAI,MAAM,aAAa,CACrB,aAAY,MAAM,SAAS,KAAK,QAAQ,EAAE;YACjC,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,CACrD,SAAQ,KAAK,KAAK;;SAGhB;AACR,QAAO;;AAGT,SAAS,iBAAiB,UAA4B;AACpD,KAAI,CAAC,WAAW,SAAS,CAAE,QAAO,EAAE;CACpC,MAAM,UAAoB,EAAE;AAC5B,KAAI;AACF,OAAK,MAAM,SAAS,YAAY,UAAU,EAAE,eAAe,MAAM,CAAC,EAAE;AAClE,OAAI,QAAQ,UAAU,sBAAuB;AAC7C,OAAI,MAAM,gBAAgB,CAAE;AAC5B,OAAI,oBAAoB,IAAI,MAAM,KAAK,CAAE;AACzC,OAAI,uBAAuB,IAAI,MAAM,KAAK,CAAE;GAC5C,MAAM,OAAO,KAAK,UAAU,MAAM,KAAK;AACvC,OAAI,MAAM,aAAa,CACrB,aAAY,MAAM,SAAS,sBAAsB;YACxC,MAAM,QAAQ,IAAI,MAAM,KAAK,SAAS,MAAM,EACrD;QAAI,MAAM,SAAS,YAAa,SAAQ,KAAK,KAAK;;;SAGhD;AACR,QAAO;;AAGT,SAAS,6BAA6B,UAA2B;CAC/D,MAAM,aAAa,UAAU,SAAS;CACtC,MAAM,OAAO,SAAS;AACtB,KAAI,KAAK,WAAW,WAAW,IAAI,eAAe,IAAK,QAAO;AAC9D,KAAI,WAAW,WAAW,KAAK,EAAE;EAC/B,MAAM,MAAM,WAAW,MAAM,KAAK,OAAO,CAAC,QAAQ,OAAO,GAAG;AAE5D,OADc,MAAM,IAAI,MAAM,IAAI,CAAC,SAAS,OAC9B,EAAG,QAAO;;AAE1B,KAAI,WAAW,KAAK,YAAY,OAAO,CAAC,CAAE,QAAO;AACjD,QAAO;;AAGT,SAAS,mBAAkC;AACzC,QAAO,IAAI,SAAS,YAAY,aAAa,QAAQ,CAAC;;;;;;AAWxD,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;CAKtF;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,aAAyD;IAC7D,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,SAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;;;AAI5C,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,aAAyD;KAC7D,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,UAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;;;AAI5C,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;AAG3C,MAFgB,MAAM,qBAAqB,SAAS,WAAW,UAAU,SAAS,QAAQ,KAAK,EAElF;GAEX,MAAM,MAAM,MAAM,QAAQ,YAAY,WAAW,QAAQ;AACzD,UAAO;AACP,UAAO,iBAAiB,IAAI;QAE5B,QAAO;;AAIX,QAAO;;AAOT,MAAM,mBAAmB;AACzB,MAAM,oBAAoB;;;;;;;;;;;;;AAc1B,eAAsB,uBACpB,SACA,YACiB;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;AAEf,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,SAAS,MAAM;AAG3B,QAAK,WAAW,KAAK,sBAAsB,EACzC,OAAM,kBAAkB;GAI1B,MAAM,OAAO,mBADD,MAAM,kBAAkB,KAAK,CACL;AACpC,SAAM,QAAQ,gBAAgB,IAAI,KAAK;;AAGzC,cAAY,MAAM;AAClB,UAAQ,OAAO,MACb,yBAAyB,SAAS,GAAG,MAAM,WAC5C;;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"}