@tekmidian/pai 0.5.7 → 0.6.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (137) hide show
  1. package/ARCHITECTURE.md +72 -1
  2. package/README.md +87 -1
  3. package/dist/{auto-route-BG6I_4B1.mjs → auto-route-C-DrW6BL.mjs} +3 -3
  4. package/dist/{auto-route-BG6I_4B1.mjs.map → auto-route-C-DrW6BL.mjs.map} +1 -1
  5. package/dist/cli/index.mjs +1482 -1628
  6. package/dist/cli/index.mjs.map +1 -1
  7. package/dist/clusters-JIDQW65f.mjs +201 -0
  8. package/dist/clusters-JIDQW65f.mjs.map +1 -0
  9. package/dist/{config-Cf92lGX_.mjs → config-BuhHWyOK.mjs} +21 -6
  10. package/dist/config-BuhHWyOK.mjs.map +1 -0
  11. package/dist/daemon/index.mjs +11 -8
  12. package/dist/daemon/index.mjs.map +1 -1
  13. package/dist/{daemon-2ND5WO2j.mjs → daemon-D3hYb5_C.mjs} +669 -218
  14. package/dist/daemon-D3hYb5_C.mjs.map +1 -0
  15. package/dist/daemon-mcp/index.mjs +4597 -4
  16. package/dist/daemon-mcp/index.mjs.map +1 -1
  17. package/dist/db-DdUperSl.mjs +110 -0
  18. package/dist/db-DdUperSl.mjs.map +1 -0
  19. package/dist/{detect-BU3Nx_2L.mjs → detect-CdaA48EI.mjs} +1 -1
  20. package/dist/{detect-BU3Nx_2L.mjs.map → detect-CdaA48EI.mjs.map} +1 -1
  21. package/dist/{detector-Bp-2SM3x.mjs → detector-jGBuYQJM.mjs} +2 -2
  22. package/dist/{detector-Bp-2SM3x.mjs.map → detector-jGBuYQJM.mjs.map} +1 -1
  23. package/dist/{factory-Bzcy70G9.mjs → factory-Ygqe_bVZ.mjs} +7 -5
  24. package/dist/{factory-Bzcy70G9.mjs.map → factory-Ygqe_bVZ.mjs.map} +1 -1
  25. package/dist/helpers-BEST-4Gx.mjs +420 -0
  26. package/dist/helpers-BEST-4Gx.mjs.map +1 -0
  27. package/dist/hooks/capture-all-events.mjs +2 -2
  28. package/dist/hooks/capture-all-events.mjs.map +3 -3
  29. package/dist/hooks/capture-session-summary.mjs +38 -0
  30. package/dist/hooks/capture-session-summary.mjs.map +3 -3
  31. package/dist/hooks/cleanup-session-files.mjs +6 -12
  32. package/dist/hooks/cleanup-session-files.mjs.map +4 -4
  33. package/dist/hooks/context-compression-hook.mjs +93 -104
  34. package/dist/hooks/context-compression-hook.mjs.map +4 -4
  35. package/dist/hooks/initialize-session.mjs +14 -11
  36. package/dist/hooks/initialize-session.mjs.map +4 -4
  37. package/dist/hooks/inject-observations.mjs +220 -0
  38. package/dist/hooks/inject-observations.mjs.map +7 -0
  39. package/dist/hooks/load-core-context.mjs +2 -2
  40. package/dist/hooks/load-core-context.mjs.map +3 -3
  41. package/dist/hooks/load-project-context.mjs +90 -91
  42. package/dist/hooks/load-project-context.mjs.map +4 -4
  43. package/dist/hooks/observe.mjs +354 -0
  44. package/dist/hooks/observe.mjs.map +7 -0
  45. package/dist/hooks/stop-hook.mjs +94 -107
  46. package/dist/hooks/stop-hook.mjs.map +4 -4
  47. package/dist/hooks/sync-todo-to-md.mjs +31 -33
  48. package/dist/hooks/sync-todo-to-md.mjs.map +4 -4
  49. package/dist/index.d.mts +30 -7
  50. package/dist/index.d.mts.map +1 -1
  51. package/dist/index.mjs +5 -8
  52. package/dist/indexer-D53l5d1U.mjs +1 -0
  53. package/dist/{indexer-backend-CIMXedqk.mjs → indexer-backend-jcJFsmB4.mjs} +37 -127
  54. package/dist/indexer-backend-jcJFsmB4.mjs.map +1 -0
  55. package/dist/{ipc-client-Bjg_a1dc.mjs → ipc-client-CoyUHPod.mjs} +2 -7
  56. package/dist/{ipc-client-Bjg_a1dc.mjs.map → ipc-client-CoyUHPod.mjs.map} +1 -1
  57. package/dist/latent-ideas-bTJo6Omd.mjs +191 -0
  58. package/dist/latent-ideas-bTJo6Omd.mjs.map +1 -0
  59. package/dist/neighborhood-BYYbEkUJ.mjs +135 -0
  60. package/dist/neighborhood-BYYbEkUJ.mjs.map +1 -0
  61. package/dist/note-context-BK24bX8Y.mjs +126 -0
  62. package/dist/note-context-BK24bX8Y.mjs.map +1 -0
  63. package/dist/postgres-CKf-EDtS.mjs +846 -0
  64. package/dist/postgres-CKf-EDtS.mjs.map +1 -0
  65. package/dist/{reranker-D7bRAHi6.mjs → reranker-CMNZcfVx.mjs} +1 -1
  66. package/dist/{reranker-D7bRAHi6.mjs.map → reranker-CMNZcfVx.mjs.map} +1 -1
  67. package/dist/{search-_oHfguA5.mjs → search-DC1qhkKn.mjs} +2 -58
  68. package/dist/search-DC1qhkKn.mjs.map +1 -0
  69. package/dist/{sqlite-WWBq7_2C.mjs → sqlite-l-s9xPjY.mjs} +160 -3
  70. package/dist/sqlite-l-s9xPjY.mjs.map +1 -0
  71. package/dist/state-C6_vqz7w.mjs +102 -0
  72. package/dist/state-C6_vqz7w.mjs.map +1 -0
  73. package/dist/stop-words-BaMEGVeY.mjs +326 -0
  74. package/dist/stop-words-BaMEGVeY.mjs.map +1 -0
  75. package/dist/{indexer-CMPOiY1r.mjs → sync-BOsnEj2-.mjs} +14 -216
  76. package/dist/sync-BOsnEj2-.mjs.map +1 -0
  77. package/dist/themes-BvYF0W8T.mjs +148 -0
  78. package/dist/themes-BvYF0W8T.mjs.map +1 -0
  79. package/dist/{tools-DV_lsiCc.mjs → tools-DcaJlYDN.mjs} +162 -273
  80. package/dist/tools-DcaJlYDN.mjs.map +1 -0
  81. package/dist/trace-CRx9lPuc.mjs +137 -0
  82. package/dist/trace-CRx9lPuc.mjs.map +1 -0
  83. package/dist/{vault-indexer-k-kUlaZ-.mjs → vault-indexer-Bi2cRmn7.mjs} +134 -132
  84. package/dist/vault-indexer-Bi2cRmn7.mjs.map +1 -0
  85. package/dist/zettelkasten-cdajbnPr.mjs +708 -0
  86. package/dist/zettelkasten-cdajbnPr.mjs.map +1 -0
  87. package/package.json +1 -2
  88. package/src/hooks/ts/lib/project-utils/index.ts +50 -0
  89. package/src/hooks/ts/lib/project-utils/notify.ts +75 -0
  90. package/src/hooks/ts/lib/project-utils/paths.ts +218 -0
  91. package/src/hooks/ts/lib/project-utils/session-notes.ts +363 -0
  92. package/src/hooks/ts/lib/project-utils/todo.ts +178 -0
  93. package/src/hooks/ts/lib/project-utils/tokens.ts +39 -0
  94. package/src/hooks/ts/lib/project-utils.ts +40 -1018
  95. package/src/hooks/ts/post-tool-use/observe.ts +327 -0
  96. package/src/hooks/ts/session-end/capture-session-summary.ts +41 -0
  97. package/src/hooks/ts/session-start/inject-observations.ts +254 -0
  98. package/dist/chunker-CbnBe0s0.mjs +0 -191
  99. package/dist/chunker-CbnBe0s0.mjs.map +0 -1
  100. package/dist/config-Cf92lGX_.mjs.map +0 -1
  101. package/dist/daemon-2ND5WO2j.mjs.map +0 -1
  102. package/dist/db-Dp8VXIMR.mjs +0 -212
  103. package/dist/db-Dp8VXIMR.mjs.map +0 -1
  104. package/dist/indexer-CMPOiY1r.mjs.map +0 -1
  105. package/dist/indexer-backend-CIMXedqk.mjs.map +0 -1
  106. package/dist/mcp/index.d.mts +0 -1
  107. package/dist/mcp/index.mjs +0 -500
  108. package/dist/mcp/index.mjs.map +0 -1
  109. package/dist/postgres-FXrHDPcE.mjs +0 -358
  110. package/dist/postgres-FXrHDPcE.mjs.map +0 -1
  111. package/dist/schemas-BFIgGntb.mjs +0 -3405
  112. package/dist/schemas-BFIgGntb.mjs.map +0 -1
  113. package/dist/search-_oHfguA5.mjs.map +0 -1
  114. package/dist/sqlite-WWBq7_2C.mjs.map +0 -1
  115. package/dist/tools-DV_lsiCc.mjs.map +0 -1
  116. package/dist/vault-indexer-k-kUlaZ-.mjs.map +0 -1
  117. package/dist/zettelkasten-e-a4rW_6.mjs +0 -901
  118. package/dist/zettelkasten-e-a4rW_6.mjs.map +0 -1
  119. package/templates/README.md +0 -181
  120. package/templates/skills/CORE/Aesthetic.md +0 -333
  121. package/templates/skills/CORE/CONSTITUTION.md +0 -1502
  122. package/templates/skills/CORE/HistorySystem.md +0 -427
  123. package/templates/skills/CORE/HookSystem.md +0 -1082
  124. package/templates/skills/CORE/Prompting.md +0 -509
  125. package/templates/skills/CORE/ProsodyAgentTemplate.md +0 -53
  126. package/templates/skills/CORE/ProsodyGuide.md +0 -416
  127. package/templates/skills/CORE/SKILL.md +0 -741
  128. package/templates/skills/CORE/SkillSystem.md +0 -213
  129. package/templates/skills/CORE/TerminalTabs.md +0 -119
  130. package/templates/skills/CORE/VOICE.md +0 -106
  131. package/templates/skills/createskill-skill.template.md +0 -78
  132. package/templates/skills/history-system.template.md +0 -371
  133. package/templates/skills/hook-system.template.md +0 -913
  134. package/templates/skills/sessions-skill.template.md +0 -102
  135. package/templates/skills/skill-system.template.md +0 -214
  136. package/templates/skills/terminal-tabs.template.md +0 -120
  137. package/templates/templates.md +0 -20
@@ -1 +0,0 @@
1
- {"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"}
@@ -1 +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"}
@@ -1 +0,0 @@
1
- export { };
@@ -1,500 +0,0 @@
1
- #!/usr/bin/env node
2
- import { n as openRegistry } from "../db-BtuN768f.mjs";
3
- import { n as openFederation } from "../db-Dp8VXIMR.mjs";
4
- import "../embeddings-DGRAPAYb.mjs";
5
- import "../search-_oHfguA5.mjs";
6
- import { a as record, i as number, n as array, o as string, r as boolean, s as unknown, t as _enum } from "../schemas-BFIgGntb.mjs";
7
- import { o as loadConfig } from "../config-Cf92lGX_.mjs";
8
- import { _ as toolZettelSurprise, a as toolProjectHealth, d as toolSessionRoute, f as toolTopicDetect, g as toolZettelSuggest, h as toolZettelHealth, i as toolProjectDetect, l as toolRegistrySearch, m as toolZettelExplore, n as toolMemorySearch, o as toolProjectInfo, p as toolZettelConverse, r as toolNotificationConfig, s as toolProjectList, t as toolMemoryGet, u as toolSessionList, v as toolZettelThemes } from "../tools-DV_lsiCc.mjs";
9
- import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
10
- import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
11
-
12
- //#region src/mcp/server.ts
13
- /**
14
- * PAI Knowledge OS — MCP Server (Phase 3)
15
- *
16
- * Exposes PAI registry and memory as MCP tools callable by Claude Code.
17
- *
18
- * Tools:
19
- * memory_search — BM25 search across indexed memory/notes
20
- * memory_get — Read a specific file or lines from a project
21
- * project_info — Get details for a project (by slug or current dir)
22
- * project_list — List projects with optional filters
23
- * session_list — List sessions for a project
24
- * registry_search — Full-text search over project slugs/names/paths
25
- * project_detect — Detect which project a path belongs to
26
- * project_health — Audit all projects for moved/deleted directories
27
- * session_route — Auto-route session to project (path/marker/topic)
28
- *
29
- * NOTE: All tool logic lives in tools.ts (shared with the daemon).
30
- * This file wires MCP schema definitions to those pure functions.
31
- */
32
- let _registryDb = null;
33
- let _federationDb = null;
34
- function getRegistryDb() {
35
- if (!_registryDb) _registryDb = openRegistry();
36
- return _registryDb;
37
- }
38
- function getFederationDb() {
39
- if (!_federationDb) _federationDb = openFederation();
40
- return _federationDb;
41
- }
42
- async function startMcpServer() {
43
- const server = new McpServer({
44
- name: "pai",
45
- version: "0.1.0"
46
- });
47
- server.tool("memory_search", [
48
- "Search PAI federated memory using BM25 full-text ranking, semantic similarity, or a hybrid of both.",
49
- "",
50
- "Use this BEFORE answering questions about past work, decisions, dates, people,",
51
- "preferences, project status, todos, technical choices, or anything that might",
52
- "have been recorded in session notes or memory files.",
53
- "",
54
- "Modes:",
55
- " keyword — BM25 full-text search (default, fast, no embeddings required)",
56
- " semantic — Cosine similarity over vector embeddings (requires prior embed run)",
57
- " hybrid — Normalized combination of BM25 + cosine (best quality)",
58
- "",
59
- "Reranking is ON by default — results are re-scored with a cross-encoder model for better relevance.",
60
- "Set rerank=false to skip reranking (faster but less accurate ordering).",
61
- "",
62
- "Recency boost optionally down-weights older results (recency_boost=90 means scores halve every 90 days).",
63
- "",
64
- "Defaults come from ~/.config/pai/config.json (search section). Per-call parameters override config defaults.",
65
- "",
66
- "Returns ranked snippets with project slug, file path, line range, and score.",
67
- "Higher score = more relevant."
68
- ].join("\n"), {
69
- query: string().describe("Free-text search query. Multiple words are ORed together — any matching word returns a result, ranked by relevance."),
70
- project: string().optional().describe("Scope search to a single project by slug. Omit to search all projects."),
71
- all_projects: boolean().optional().describe("Explicitly search all projects (default behaviour when project is omitted)."),
72
- sources: array(_enum(["memory", "notes"])).optional().describe("Restrict to specific source types: 'memory' or 'notes'."),
73
- limit: number().int().min(1).max(100).optional().describe("Maximum results to return. Default: 10."),
74
- mode: _enum([
75
- "keyword",
76
- "semantic",
77
- "hybrid"
78
- ]).optional().describe("Search mode: 'keyword' (BM25, default), 'semantic' (vector cosine), or 'hybrid' (both combined)."),
79
- rerank: boolean().optional().describe("Rerank results using a cross-encoder model for better relevance. Default: true. Set to false to skip reranking for faster but less accurate results."),
80
- recency_boost: number().int().min(0).max(365).optional().describe("Apply recency boost: score halves every N days. 0 = off. Default from config (typically 90). Applied after reranking.")
81
- }, async (args) => {
82
- const config = loadConfig();
83
- const result = await toolMemorySearch(getRegistryDb(), getFederationDb(), {
84
- ...args,
85
- recencyBoost: args.recency_boost
86
- }, config.search);
87
- return {
88
- content: result.content.map((c) => ({
89
- type: c.type,
90
- text: c.text
91
- })),
92
- isError: result.isError
93
- };
94
- });
95
- server.tool("memory_get", [
96
- "Read the content of a specific file from a registered PAI project.",
97
- "",
98
- "Use this to read a full memory file, session note, or document after finding",
99
- "it via memory_search. Optionally restrict to a line range.",
100
- "",
101
- "The path must be a relative path within the project root (no ../ traversal)."
102
- ].join("\n"), {
103
- project: string().describe("Project slug identifying which project's files to read from."),
104
- path: string().describe("Relative path within the project root (e.g. 'Notes/0001 - 2026-01-01 - Example.md')."),
105
- from: number().int().min(1).optional().describe("Starting line number (1-based, inclusive). Default: 1."),
106
- lines: number().int().min(1).optional().describe("Number of lines to return. Default: entire file.")
107
- }, async (args) => {
108
- const result = toolMemoryGet(getRegistryDb(), args);
109
- return {
110
- content: result.content.map((c) => ({
111
- type: c.type,
112
- text: c.text
113
- })),
114
- isError: result.isError
115
- };
116
- });
117
- server.tool("project_info", [
118
- "Get detailed information about a PAI registered project.",
119
- "",
120
- "Use this to look up a project's root path, type, status, tags, session count,",
121
- "and last active date. If no slug is provided, attempts to detect the current",
122
- "project from the caller's working directory."
123
- ].join("\n"), { slug: string().optional().describe("Project slug. Omit to auto-detect from the current working directory.") }, async (args) => {
124
- const result = toolProjectInfo(getRegistryDb(), args);
125
- return {
126
- content: result.content.map((c) => ({
127
- type: c.type,
128
- text: c.text
129
- })),
130
- isError: result.isError
131
- };
132
- });
133
- server.tool("project_list", [
134
- "List registered PAI projects with optional filters.",
135
- "",
136
- "Use this to browse all known projects, find projects by status or tag,",
137
- "or get a quick overview of the PAI registry."
138
- ].join("\n"), {
139
- status: _enum([
140
- "active",
141
- "archived",
142
- "migrating"
143
- ]).optional().describe("Filter by project status. Default: all statuses."),
144
- tag: string().optional().describe("Filter by tag name (exact match)."),
145
- limit: number().int().min(1).max(500).optional().describe("Maximum number of projects to return. Default: 50.")
146
- }, async (args) => {
147
- const result = toolProjectList(getRegistryDb(), args);
148
- return {
149
- content: result.content.map((c) => ({
150
- type: c.type,
151
- text: c.text
152
- })),
153
- isError: result.isError
154
- };
155
- });
156
- server.tool("session_list", [
157
- "List session notes for a PAI project.",
158
- "",
159
- "Use this to find what sessions exist for a project, see their dates and titles,",
160
- "and identify specific session notes to read via memory_get."
161
- ].join("\n"), {
162
- project: string().describe("Project slug to list sessions for."),
163
- limit: number().int().min(1).max(500).optional().describe("Maximum sessions to return. Default: 10 (most recent first)."),
164
- status: _enum([
165
- "open",
166
- "completed",
167
- "compacted"
168
- ]).optional().describe("Filter by session status.")
169
- }, async (args) => {
170
- const result = toolSessionList(getRegistryDb(), args);
171
- return {
172
- content: result.content.map((c) => ({
173
- type: c.type,
174
- text: c.text
175
- })),
176
- isError: result.isError
177
- };
178
- });
179
- server.tool("registry_search", [
180
- "Search PAI project registry by slug, display name, or path.",
181
- "",
182
- "Use this to find the slug for a project when you know its name or path,",
183
- "or to check if a project is registered. Returns matching project entries."
184
- ].join("\n"), { query: string().describe("Search term matched against project slugs, display names, and root paths (case-insensitive substring match).") }, async (args) => {
185
- const result = toolRegistrySearch(getRegistryDb(), args);
186
- return {
187
- content: result.content.map((c) => ({
188
- type: c.type,
189
- text: c.text
190
- })),
191
- isError: result.isError
192
- };
193
- });
194
- server.tool("project_detect", [
195
- "Detect which registered PAI project a filesystem path belongs to.",
196
- "",
197
- "Use this at session start to auto-identify the current project from the",
198
- "working directory, or to map any path back to its registered project.",
199
- "",
200
- "Returns: slug, display_name, root_path, type, status, match_type (exact|parent),",
201
- "relative_path (if the given path is inside a project), and session stats.",
202
- "",
203
- "match_type 'exact' means the path IS the project root.",
204
- "match_type 'parent' means the path is a subdirectory of the project root."
205
- ].join("\n"), { cwd: string().optional().describe("Absolute path to detect project for. Defaults to the MCP server's process.cwd().") }, async (args) => {
206
- const result = toolProjectDetect(getRegistryDb(), args);
207
- return {
208
- content: result.content.map((c) => ({
209
- type: c.type,
210
- text: c.text
211
- })),
212
- isError: result.isError
213
- };
214
- });
215
- server.tool("project_health", [
216
- "Audit all registered PAI projects to find moved or deleted directories.",
217
- "",
218
- "Returns a JSON report categorising every project as:",
219
- " active — root_path exists on disk",
220
- " stale — root_path missing, but a directory with the same name was found nearby",
221
- " dead — root_path missing, no candidate found",
222
- "",
223
- "Use this to diagnose orphaned sessions or missing project paths."
224
- ].join("\n"), { category: _enum([
225
- "active",
226
- "stale",
227
- "dead",
228
- "all"
229
- ]).optional().describe("Filter results to a specific health category. Default: all.") }, async (args) => {
230
- const result = await toolProjectHealth(getRegistryDb(), args);
231
- return {
232
- content: result.content.map((c) => ({
233
- type: c.type,
234
- text: c.text
235
- })),
236
- isError: result.isError
237
- };
238
- });
239
- server.tool("notification_config", [
240
- "Query or update the PAI unified notification configuration.",
241
- "",
242
- "Actions:",
243
- " get — Return the current notification mode, active channels, and routing table.",
244
- " set — Change the notification mode or update channel/routing config.",
245
- " send — Send a notification through the configured channels.",
246
- "",
247
- "Notification modes:",
248
- " auto — Use the per-event routing table (default)",
249
- " voice — All events sent as WhatsApp voice (TTS)",
250
- " whatsapp — All events sent as WhatsApp text",
251
- " ntfy — All events sent to ntfy.sh",
252
- " macos — All events sent as macOS notifications",
253
- " cli — All events written to CLI output only",
254
- " off — Suppress all notifications",
255
- "",
256
- "Event types for send: error | progress | completion | info | debug",
257
- "",
258
- "Examples:",
259
- " { \"action\": \"get\" }",
260
- " { \"action\": \"set\", \"mode\": \"voice\" }",
261
- " { \"action\": \"send\", \"event\": \"completion\", \"message\": \"Done!\" }"
262
- ].join("\n"), {
263
- action: _enum([
264
- "get",
265
- "set",
266
- "send"
267
- ]).describe("Action: 'get' (read config), 'set' (update config), 'send' (send notification)."),
268
- mode: _enum([
269
- "auto",
270
- "voice",
271
- "whatsapp",
272
- "ntfy",
273
- "macos",
274
- "cli",
275
- "off"
276
- ]).optional().describe("For action=set: new notification mode."),
277
- channels: record(string(), unknown()).optional().describe("For action=set: partial channel config overrides as a JSON object. E.g. { \"whatsapp\": { \"enabled\": true }, \"macos\": { \"enabled\": false } }"),
278
- routing: record(string(), unknown()).optional().describe("For action=set: partial routing overrides as a JSON object. E.g. { \"error\": [\"whatsapp\", \"macos\"], \"progress\": [\"cli\"] }"),
279
- event: _enum([
280
- "error",
281
- "progress",
282
- "completion",
283
- "info",
284
- "debug"
285
- ]).optional().describe("For action=send: event type. Default: 'info'."),
286
- message: string().optional().describe("For action=send: the notification message body."),
287
- title: string().optional().describe("For action=send: optional notification title (used by macOS and ntfy).")
288
- }, async (args) => {
289
- const result = await toolNotificationConfig(args);
290
- return {
291
- content: result.content.map((c) => ({
292
- type: c.type,
293
- text: c.text
294
- })),
295
- isError: result.isError
296
- };
297
- });
298
- server.tool("topic_detect", [
299
- "Detect whether recent conversation context has shifted to a different project.",
300
- "",
301
- "Call this when the conversation may have drifted away from the initially-routed project.",
302
- "Provide a short summary of the recent context (last few messages or tool call results).",
303
- "",
304
- "Returns:",
305
- " shifted — true if a topic shift was detected",
306
- " current_project — the project the session is currently routed to",
307
- " suggested_project — the project that best matches the context",
308
- " confidence — [0,1] fraction of memory mass held by suggested_project",
309
- " chunks_scored — number of memory chunks that contributed to scoring",
310
- " top_matches — top-3 projects with their confidence percentages",
311
- "",
312
- "A shift is reported when confidence >= threshold (default 0.6) and the",
313
- "best-matching project differs from current_project.",
314
- "",
315
- "Use cases:",
316
- " - Call at session start to confirm routing is correct",
317
- " - Call periodically when working across multiple concerns",
318
- " - Integrate with pre-tool hooks for automatic drift detection"
319
- ].join("\n"), {
320
- context: string().describe("Recent conversation context: a few sentences summarising what the session has been discussing. Can include file paths, feature names, commands run, or any relevant text."),
321
- current_project: string().optional().describe("The project slug this session is currently routed to. If omitted, the tool still returns the best-matching project but shifted will always be false."),
322
- threshold: number().min(0).max(1).optional().describe("Minimum confidence [0,1] to declare a shift. Default: 0.6. Increase to reduce false positives. Decrease to catch subtle drifts.")
323
- }, async (args) => {
324
- const result = await toolTopicDetect(args);
325
- return {
326
- content: result.content.map((c) => ({
327
- type: c.type,
328
- text: c.text
329
- })),
330
- isError: result.isError
331
- };
332
- });
333
- server.tool("session_route", [
334
- "Automatically detect which project this session belongs to.",
335
- "",
336
- "Call this at session start (e.g., from CLAUDE.md or a session-start hook)",
337
- "to route the session to the correct project automatically.",
338
- "",
339
- "Detection strategy (in priority order):",
340
- " 1. path — exact or parent-directory match in the project registry",
341
- " 2. marker — walk up from cwd looking for Notes/PAI.md marker files",
342
- " 3. topic — BM25 keyword search against memory (only if context provided)",
343
- "",
344
- "Returns:",
345
- " slug — the matched project slug",
346
- " display_name — human-readable project name",
347
- " root_path — absolute path to the project root",
348
- " method — how it was detected: 'path', 'marker', or 'topic'",
349
- " confidence — 1.0 for path/marker matches, BM25 fraction for topic",
350
- "",
351
- "If no match is found, returns a message explaining what was tried.",
352
- "Run 'pai project add .' to register the current directory."
353
- ].join("\n"), {
354
- cwd: string().optional().describe("Working directory to detect from. Defaults to process.cwd(). Pass the session's actual working directory for accurate detection."),
355
- context: string().optional().describe("Optional conversation context for topic-based fallback routing. A few sentences summarising what the session will work on. Only used if path and marker detection both fail.")
356
- }, async (args) => {
357
- const result = await toolSessionRoute(getRegistryDb(), getFederationDb(), args);
358
- return {
359
- content: result.content.map((c) => ({
360
- type: c.type,
361
- text: c.text
362
- })),
363
- isError: result.isError
364
- };
365
- });
366
- server.tool("zettel_explore", [
367
- "Explore the vault's knowledge graph using Luhmann's Folgezettel traversal.",
368
- "Follow trains of thought forward, backward, or both from a starting note.",
369
- "Classifies links as sequential (same-folder) or associative (cross-folder)."
370
- ].join("\n"), {
371
- start_note: string().describe("Path or title of the note to start traversal from."),
372
- depth: number().int().min(1).max(10).optional().describe("How many link hops to traverse. Default: 3."),
373
- direction: _enum([
374
- "forward",
375
- "backward",
376
- "both"
377
- ]).optional().describe("Traversal direction: 'forward' (outlinks), 'backward' (backlinks), or 'both'. Default: both."),
378
- mode: _enum([
379
- "sequential",
380
- "associative",
381
- "all"
382
- ]).optional().describe("Link type filter: 'sequential' (same-folder), 'associative' (cross-folder), or 'all'. Default: all.")
383
- }, async (args) => {
384
- const result = await toolZettelExplore(getFederationDb(), args);
385
- return {
386
- content: result.content.map((c) => ({
387
- type: c.type,
388
- text: c.text
389
- })),
390
- isError: result.isError
391
- };
392
- });
393
- server.tool("zettel_health", ["Audit the structural health of the Obsidian vault.", "Reports dead links, orphan notes, disconnected clusters, low-connectivity files, and an overall health score."].join("\n"), {
394
- scope: _enum([
395
- "full",
396
- "recent",
397
- "project"
398
- ]).optional().describe("Audit scope: 'full' (entire vault), 'recent' (recently modified), or 'project' (specific path). Default: full."),
399
- project_path: string().optional().describe("Absolute path to the project/folder to audit when scope='project'."),
400
- recent_days: number().int().optional().describe("Number of days to look back when scope='recent'. Default: 30."),
401
- include: array(_enum([
402
- "dead_links",
403
- "orphans",
404
- "disconnected",
405
- "low_connectivity"
406
- ])).optional().describe("Specific checks to include. Omit to run all checks.")
407
- }, async (args) => {
408
- const result = await toolZettelHealth(getFederationDb(), args);
409
- return {
410
- content: result.content.map((c) => ({
411
- type: c.type,
412
- text: c.text
413
- })),
414
- isError: result.isError
415
- };
416
- });
417
- server.tool("zettel_surprise", ["Find surprising connections — notes that are semantically similar to a reference note but far away in the link graph.", "High surprise = unexpected relevance."].join("\n"), {
418
- reference_path: string().describe("Path to the reference note to find surprising connections for."),
419
- vault_project_id: number().int().describe("Project ID of the vault to search within."),
420
- limit: number().int().optional().describe("Maximum number of surprising notes to return. Default: 10."),
421
- min_similarity: number().optional().describe("Minimum semantic similarity [0,1] for a note to be considered. Default: 0.5."),
422
- min_graph_distance: number().int().optional().describe("Minimum link hops away from the reference note. Default: 3.")
423
- }, async (args) => {
424
- const result = await toolZettelSurprise(getFederationDb(), args);
425
- return {
426
- content: result.content.map((c) => ({
427
- type: c.type,
428
- text: c.text
429
- })),
430
- isError: result.isError
431
- };
432
- });
433
- server.tool("zettel_suggest", ["Suggest new connections for a note using semantic similarity, shared tags, and graph neighborhood (friends-of-friends)."].join("\n"), {
434
- note_path: string().describe("Path to the note to generate link suggestions for."),
435
- vault_project_id: number().int().describe("Project ID of the vault to search within."),
436
- limit: number().int().optional().describe("Maximum number of suggestions to return. Default: 10."),
437
- exclude_linked: boolean().optional().describe("Exclude notes already linked from this note. Default: true.")
438
- }, async (args) => {
439
- const result = await toolZettelSuggest(getFederationDb(), args);
440
- return {
441
- content: result.content.map((c) => ({
442
- type: c.type,
443
- text: c.text
444
- })),
445
- isError: result.isError
446
- };
447
- });
448
- server.tool("zettel_converse", ["Use the vault as a Zettelkasten communication partner.", "Ask a question, get relevant notes with cross-domain connections and a synthesis prompt for generating new insights."].join("\n"), {
449
- question: string().describe("The question or topic to explore in the vault."),
450
- vault_project_id: number().int().describe("Project ID of the vault to query."),
451
- depth: number().int().optional().describe("How many link hops to follow from seed notes. Default: 2."),
452
- limit: number().int().optional().describe("Maximum number of relevant notes to retrieve. Default: 10.")
453
- }, async (args) => {
454
- const result = await toolZettelConverse(getFederationDb(), args);
455
- return {
456
- content: result.content.map((c) => ({
457
- type: c.type,
458
- text: c.text
459
- })),
460
- isError: result.isError
461
- };
462
- });
463
- server.tool("zettel_themes", ["Detect emerging themes by clustering recent notes with similar embeddings.", "Reveals forming idea clusters and suggests index notes for unlinked clusters."].join("\n"), {
464
- vault_project_id: number().int().describe("Project ID of the vault to analyse."),
465
- lookback_days: number().int().optional().describe("Number of days of recent notes to cluster. Default: 30."),
466
- min_cluster_size: number().int().optional().describe("Minimum notes required to form a theme cluster. Default: 3."),
467
- max_themes: number().int().optional().describe("Maximum number of theme clusters to return. Default: 10."),
468
- similarity_threshold: number().optional().describe("Minimum cosine similarity to group notes into a cluster [0,1]. Default: 0.7.")
469
- }, async (args) => {
470
- const result = await toolZettelThemes(getFederationDb(), args);
471
- return {
472
- content: result.content.map((c) => ({
473
- type: c.type,
474
- text: c.text
475
- })),
476
- isError: result.isError
477
- };
478
- });
479
- const transport = new StdioServerTransport();
480
- await server.connect(transport);
481
- }
482
-
483
- //#endregion
484
- //#region src/mcp/index.ts
485
- /**
486
- * PAI Knowledge OS — MCP server entry point
487
- *
488
- * When invoked as `node dist/mcp/index.mjs` (or via the `pai-mcp` bin),
489
- * starts the PAI MCP server on stdio transport so Claude Code can call
490
- * memory_search, memory_get, project_info, project_list, session_list,
491
- * and registry_search tools directly during conversations.
492
- */
493
- startMcpServer().catch((err) => {
494
- process.stderr.write(`PAI MCP server fatal error: ${String(err)}\n`);
495
- process.exit(1);
496
- });
497
-
498
- //#endregion
499
- export { };
500
- //# sourceMappingURL=index.mjs.map