@tekmidian/pai 0.2.2 → 0.3.1

package/dist/{indexer-B20bPHL-.mjs → indexer-CKQcgKsz.mjs}

@@ -1,195 +1,9 @@
+import { t as chunkMarkdown } from "./chunker-CbnBe0s0.mjs";
 import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
 import { homedir } from "node:os";
 import { basename, join, normalize, relative } from "node:path";
 import { createHash } from "node:crypto";
 
-//#region src/memory/chunker.ts
-/**
-* Markdown text chunker for the PAI memory engine.
-*
-* Splits markdown files into overlapping text segments suitable for BM25
-* full-text indexing.  Respects heading boundaries where possible, falling
-* back to paragraph and sentence splitting when sections are large.
-*/
-const DEFAULT_MAX_TOKENS = 400;
-const DEFAULT_OVERLAP = 80;
-/**
-* Approximate token count using a words * 1.3 heuristic.
-* Matches the OpenClaw estimate approach.
-*/
-function estimateTokens(text) {
-	const wordCount = text.split(/\s+/).filter(Boolean).length;
-	return Math.ceil(wordCount * 1.3);
-}
-/**
-* Compute SHA-256 hash of a string, returning a hex string.
-*/
-function sha256(text) {
-	return createHash("sha256").update(text).digest("hex");
-}
-/**
-* Split content into sections delimited by ## or ### headings.
-* Each section starts at its heading line (or at line 1 for a preamble).
-*/
-function splitBySections(lines) {
-	const sections = [];
-	let current = [];
-	for (const line of lines) {
-		if (/^#{1,3}\s/.test(line.text) && current.length > 0) {
-			const text = current.map((l) => l.text).join("\n");
-			sections.push({
-				lines: current,
-				tokens: estimateTokens(text)
-			});
-			current = [];
-		}
-		current.push(line);
-	}
-	if (current.length > 0) {
-		const text = current.map((l) => l.text).join("\n");
-		sections.push({
-			lines: current,
-			tokens: estimateTokens(text)
-		});
-	}
-	return sections;
-}
-/**
-* Split a LineBlock by double-newline paragraph boundaries.
-*/
-function splitByParagraphs(block) {
-	const paragraphs = [];
-	let current = [];
-	for (const line of block.lines) if (line.text.trim() === "" && current.length > 0) {
-		const text = current.map((l) => l.text).join("\n");
-		paragraphs.push({
-			lines: [...current],
-			tokens: estimateTokens(text)
-		});
-		current = [];
-	} else current.push(line);
-	if (current.length > 0) {
-		const text = current.map((l) => l.text).join("\n");
-		paragraphs.push({
-			lines: current,
-			tokens: estimateTokens(text)
-		});
-	}
-	return paragraphs.length > 0 ? paragraphs : [block];
-}
-/**
-* Split a LineBlock by sentence boundaries (. ! ?) when even paragraphs are
-* too large.  Works character-by-character within joined lines.
-*/
-function splitBySentences(block, maxTokens) {
-	const sentences = block.lines.map((l) => l.text).join(" ").split(/(?<=[.!?])\s+(?=[A-Z"'])/g);
-	const result = [];
-	let accText = "";
-	const startLine = block.lines[0]?.lineNo ?? 1;
-	const endLine = block.lines[block.lines.length - 1]?.lineNo ?? startLine;
-	const totalLines = endLine - startLine + 1;
-	const linesPerSentence = Math.max(1, Math.floor(totalLines / Math.max(1, sentences.length)));
-	let sentenceIdx = 0;
-	let approxLine = startLine;
-	const flush = () => {
-		if (!accText.trim()) return;
-		const endApprox = Math.min(approxLine + linesPerSentence - 1, endLine);
-		result.push({
-			lines: [{
-				text: accText.trim(),
-				lineNo: approxLine
-			}],
-			tokens: estimateTokens(accText)
-		});
-		approxLine = endApprox + 1;
-		accText = "";
-	};
-	for (const sentence of sentences) {
-		sentenceIdx++;
-		const candidateText = accText ? accText + " " + sentence : sentence;
-		if (estimateTokens(candidateText) > maxTokens && accText) {
-			flush();
-			accText = sentence;
-		} else accText = candidateText;
-	}
-	flush();
-	return result.length > 0 ? result : [block];
-}
-/**
-* Extract the last `overlapTokens` worth of text from a list of previously
-* emitted chunks to prepend to the next chunk.
-*/
-function buildOverlapPrefix(chunks, overlapTokens) {
-	if (overlapTokens <= 0 || chunks.length === 0) return [];
-	const lastChunk = chunks[chunks.length - 1];
-	if (!lastChunk) return [];
-	const lines = lastChunk.text.split("\n");
-	const kept = [];
-	let acc = 0;
-	for (let i = lines.length - 1; i >= 0; i--) {
-		const lineTokens = estimateTokens(lines[i] ?? "");
-		acc += lineTokens;
-		kept.unshift(lines[i] ?? "");
-		if (acc >= overlapTokens) break;
-	}
-	const startLine = lastChunk.endLine - kept.length + 1;
-	return kept.map((text, idx) => ({
-		text,
-		lineNo: Math.max(lastChunk.startLine, startLine + idx)
-	}));
-}
-/**
-* Chunk a markdown file into overlapping segments for BM25 indexing.
-*
-* Strategy:
-*  1. Split by headings (##, ###) as natural boundaries.
-*  2. If a section exceeds maxTokens, split by paragraphs.
-*  3. If a paragraph still exceeds maxTokens, split by sentences.
-*  4. Apply overlap: each chunk includes the last `overlap` tokens from the
-*     previous chunk.
-*/
-function chunkMarkdown(content, opts) {
-	const maxTokens = opts?.maxTokens ?? DEFAULT_MAX_TOKENS;
-	const overlapTokens = opts?.overlap ?? DEFAULT_OVERLAP;
-	if (!content.trim()) return [];
-	const sections = splitBySections(content.split("\n").map((text, idx) => ({
-		text,
-		lineNo: idx + 1
-	})));
-	const finalBlocks = [];
-	for (const section of sections) {
-		if (section.tokens <= maxTokens) {
-			finalBlocks.push(section);
-			continue;
-		}
-		const paras = splitByParagraphs(section);
-		for (const para of paras) {
-			if (para.tokens <= maxTokens) {
-				finalBlocks.push(para);
-				continue;
-			}
-			const sentences = splitBySentences(para, maxTokens);
-			finalBlocks.push(...sentences);
-		}
-	}
-	const chunks = [];
-	for (const block of finalBlocks) {
-		if (block.lines.length === 0) continue;
-		const text = [...buildOverlapPrefix(chunks, overlapTokens), ...block.lines].map((l) => l.text).join("\n").trim();
-		if (!text) continue;
-		const startLine = block.lines[0]?.lineNo ?? 1;
-		const endLine = block.lines[block.lines.length - 1]?.lineNo ?? startLine;
-		chunks.push({
-			text,
-			startLine,
-			endLine,
-			hash: sha256(text)
-		});
-	}
-	return chunks;
-}
-
-//#endregion
 //#region src/memory/indexer.ts
 /**
 * File indexer for the PAI federation memory engine.
@@ -634,7 +448,7 @@ async function indexAll(db, registryDb) {
 * @param onProgress Optional callback called after each batch with running totals.
 */
 async function embedChunks(db, projectId, batchSize = 50, onProgress) {
-	const { generateEmbedding, serializeEmbedding } = await import("./embeddings-mfqv-jFu.mjs").then((n) => n.i);
+	const { generateEmbedding, serializeEmbedding } = await import("./embeddings-DGRAPAYb.mjs").then((n) => n.i);
 	const conditions = ["embedding IS NULL"];
 	const params = [];
 	if (projectId !== void 0) {
@@ -673,5 +487,5 @@ async function embedChunks(db, projectId, batchSize = 50, onProgress) {
 }
 
 //#endregion
-export { indexProject as a, indexFile as i, embedChunks as n, chunkMarkdown as o, indexAll as r, estimateTokens as s, detectTier as t };
-//# sourceMappingURL=indexer-B20bPHL-.mjs.map
+export { indexProject as a, indexFile as i, embedChunks as n, indexAll as r, detectTier as t };
+//# sourceMappingURL=indexer-CKQcgKsz.mjs.map

package/dist/indexer-CKQcgKsz.mjs.map

@@ -0,0 +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"}

package/dist/{indexer-backend-BXaocO5r.mjs → indexer-backend-BHztlJJg.mjs}

@@ -1,4 +1,5 @@
-import { o as chunkMarkdown, t as detectTier } from "./indexer-B20bPHL-.mjs";
+import { t as chunkMarkdown } from "./chunker-CbnBe0s0.mjs";
+import { t as detectTier } from "./indexer-CKQcgKsz.mjs";
 import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
 import { homedir } from "node:os";
 import { basename, join, normalize, relative } from "node:path";
@@ -313,7 +314,7 @@ const EMBED_YIELD_EVERY = 10;
 * Returns the number of newly embedded chunks.
 */
 async function embedChunksWithBackend(backend, shouldStop) {
-	const { generateEmbedding, serializeEmbedding } = await import("./embeddings-mfqv-jFu.mjs").then((n) => n.i);
+	const { generateEmbedding, serializeEmbedding } = await import("./embeddings-DGRAPAYb.mjs").then((n) => n.i);
 	const rows = await backend.getUnembeddedChunkIds();
 	if (rows.length === 0) return 0;
 	const total = rows.length;
@@ -357,4 +358,4 @@ async function indexAllWithBackend(backend, registryDb) {
 
 //#endregion
 export { embedChunksWithBackend, indexAllWithBackend };
-//# sourceMappingURL=indexer-backend-BXaocO5r.mjs.map
+//# sourceMappingURL=indexer-backend-BHztlJJg.mjs.map

package/dist/{indexer-backend-BXaocO5r.mjs.map → indexer-backend-BHztlJJg.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"indexer-backend-BXaocO5r.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"}
+{"version":3,"file":"indexer-backend-BHztlJJg.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"}

package/dist/{ipc-client-DPy7s3iu.mjs → ipc-client-CLt2fNlC.mjs}

@@ -153,4 +153,4 @@ var PaiClient = class {
 
 //#endregion
 export { ipc_client_exports as n, PaiClient as t };
-//# sourceMappingURL=ipc-client-DPy7s3iu.mjs.map
+//# sourceMappingURL=ipc-client-CLt2fNlC.mjs.map

package/dist/ipc-client-CLt2fNlC.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"ipc-client-CLt2fNlC.mjs","names":[],"sources":["../src/daemon/ipc-client.ts"],"sourcesContent":["/**\n * ipc-client.ts — IPC client for the PAI Daemon MCP shim\n *\n * PaiClient connects to the Unix Domain Socket served by daemon.ts\n * and forwards tool calls to the daemon. Uses a fresh socket connection per\n * call (connect → write JSON + newline → read response line → parse → destroy).\n * This keeps the client stateless and avoids connection management complexity.\n *\n * Adapted from the Coogle ipc-client pattern (which was adapted from Whazaa).\n */\n\nimport { connect, Socket } from \"node:net\";\nimport { randomUUID } from \"node:crypto\";\nimport type {\n  NotificationConfig,\n  NotificationMode,\n  NotificationEvent,\n  SendResult,\n} from \"../notifications/types.js\";\nimport type { TopicCheckParams, TopicCheckResult } from \"../topics/detector.js\";\nimport type { AutoRouteResult } from \"../session/auto-route.js\";\n\n// ---------------------------------------------------------------------------\n// Protocol types\n// ---------------------------------------------------------------------------\n\n/** Default socket path */\nexport const IPC_SOCKET_PATH = \"/tmp/pai.sock\";\n\n/** Timeout for IPC calls (60 seconds) */\nconst IPC_TIMEOUT_MS = 60_000;\n\ninterface IpcRequest {\n  id: string;\n  method: string;\n  params: Record<string, unknown>;\n}\n\ninterface IpcResponse {\n  id: string;\n  ok: boolean;\n  result?: unknown;\n  error?: string;\n}\n\n// ---------------------------------------------------------------------------\n// Client\n// ---------------------------------------------------------------------------\n\n/**\n * Thin IPC proxy that forwards tool calls to pai-daemon over a Unix\n * Domain Socket. Each call opens a fresh connection, sends one NDJSON request,\n * reads the response, and closes. Stateless and simple.\n */\nexport class PaiClient {\n  private readonly socketPath: string;\n\n  constructor(socketPath?: string) {\n    this.socketPath = socketPath ?? IPC_SOCKET_PATH;\n  }\n\n  /**\n   * Call a PAI tool by name with the given params.\n   * Returns the tool result or throws on error.\n   */\n  async call(method: string, params: Record<string, unknown>): Promise<unknown> {\n    return this.send(method, params);\n  }\n\n  /**\n   * Check daemon status.\n   */\n  async status(): Promise<Record<string, unknown>> {\n    const result = await this.send(\"status\", {});\n    return result as Record<string, unknown>;\n  }\n\n  /**\n   * Trigger an immediate index run.\n   */\n  async triggerIndex(): Promise<void> {\n    await this.send(\"index_now\", {});\n  }\n\n  // -------------------------------------------------------------------------\n  // Notification methods\n  // -------------------------------------------------------------------------\n\n  /**\n   * Get the current notification config from the daemon.\n   */\n  async getNotificationConfig(): Promise<{\n    config: NotificationConfig;\n    activeChannels: string[];\n  }> {\n    const result = await this.send(\"notification_get_config\", {});\n    return result as { config: NotificationConfig; activeChannels: string[] };\n  }\n\n  /**\n   * Patch the notification config on the daemon (and persist to disk).\n   */\n  async setNotificationConfig(patch: {\n    mode?: NotificationMode;\n    channels?: Partial<NotificationConfig[\"channels\"]>;\n    routing?: Partial<NotificationConfig[\"routing\"]>;\n  }): Promise<{ config: NotificationConfig }> {\n    const result = await this.send(\"notification_set_config\", patch as Record<string, unknown>);\n    return result as { config: NotificationConfig };\n  }\n\n  /**\n   * Send a notification via the daemon (routes to configured channels).\n   */\n  async sendNotification(payload: {\n    event: NotificationEvent;\n    message: string;\n    title?: string;\n  }): Promise<SendResult> {\n    const result = await this.send(\"notification_send\", payload as Record<string, unknown>);\n    return result as SendResult;\n  }\n\n  // -------------------------------------------------------------------------\n  // Topic detection methods\n  // -------------------------------------------------------------------------\n\n  /**\n   * Check whether the provided context text has drifted to a different project\n   * than the session's current routing.\n   */\n  async topicCheck(params: TopicCheckParams): Promise<TopicCheckResult> {\n    const result = await this.send(\"topic_check\", params as unknown as Record<string, unknown>);\n    return result as TopicCheckResult;\n  }\n\n  // -------------------------------------------------------------------------\n  // Session routing methods\n  // -------------------------------------------------------------------------\n\n  /**\n   * Automatically detect which project a session belongs to.\n   * Tries path match, PAI.md marker walk, then topic detection (if context given).\n   */\n  async sessionAutoRoute(params: {\n    cwd?: string;\n    context?: string;\n  }): Promise<AutoRouteResult | null> {\n    // session_auto_route returns a ToolResult (content array). Extract the text\n    // and parse JSON from it.\n    const result = await this.send(\"session_auto_route\", params as Record<string, unknown>);\n    const toolResult = result as { content?: Array<{ text: string }>; isError?: boolean };\n    if (toolResult.isError) return null;\n    const text = toolResult.content?.[0]?.text ?? \"\";\n    // Text is either JSON (on match) or a human-readable \"no match\" message\n    try {\n      return JSON.parse(text) as AutoRouteResult;\n    } catch {\n      return null;\n    }\n  }\n\n  // -------------------------------------------------------------------------\n  // Internal transport\n  // -------------------------------------------------------------------------\n\n  /**\n   * Send a single IPC request and wait for the response.\n   * Opens a new socket connection per call — simple and reliable.\n   */\n  private send(\n    method: string,\n    params: Record<string, unknown>\n  ): Promise<unknown> {\n    const socketPath = this.socketPath;\n\n    return new Promise((resolve, reject) => {\n      let socket: Socket | null = null;\n      let done = false;\n      let buffer = \"\";\n      let timer: ReturnType<typeof setTimeout> | null = null;\n\n      function finish(error: Error | null, value?: unknown): void {\n        if (done) return;\n        done = true;\n        if (timer !== null) {\n          clearTimeout(timer);\n          timer = null;\n        }\n        try {\n          socket?.destroy();\n        } catch {\n          // ignore\n        }\n        if (error) {\n          reject(error);\n        } else {\n          resolve(value);\n        }\n      }\n\n      socket = connect(socketPath, () => {\n        const request: IpcRequest = {\n          id: randomUUID(),\n          method,\n          params,\n        };\n        socket!.write(JSON.stringify(request) + \"\\n\");\n      });\n\n      socket.on(\"data\", (chunk: Buffer) => {\n        buffer += chunk.toString();\n        const nl = buffer.indexOf(\"\\n\");\n        if (nl === -1) return;\n\n        const line = buffer.slice(0, nl);\n        buffer = buffer.slice(nl + 1);\n\n        let response: IpcResponse;\n        try {\n          response = JSON.parse(line) as IpcResponse;\n        } catch {\n          finish(new Error(`IPC parse error: ${line}`));\n          return;\n        }\n\n        if (!response.ok) {\n          finish(new Error(response.error ?? \"IPC call failed\"));\n        } else {\n          finish(null, response.result);\n        }\n      });\n\n      socket.on(\"error\", (e: NodeJS.ErrnoException) => {\n        if (e.code === \"ENOENT\" || e.code === \"ECONNREFUSED\") {\n          finish(\n            new Error(\n              \"PAI daemon not running. Start it with: pai daemon serve\"\n            )\n          );\n        } else {\n          finish(e);\n        }\n      });\n\n      socket.on(\"end\", () => {\n        if (!done) {\n          finish(new Error(\"IPC connection closed before response\"));\n        }\n      });\n\n      timer = setTimeout(() => {\n        finish(new Error(\"IPC call timed out after 60s\"));\n      }, IPC_TIMEOUT_MS);\n    });\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AA2BA,MAAa,kBAAkB;;AAG/B,MAAM,iBAAiB;;;;;;AAwBvB,IAAa,YAAb,MAAuB;CACrB,AAAiB;CAEjB,YAAY,YAAqB;AAC/B,OAAK,aAAa,cAAc;;;;;;CAOlC,MAAM,KAAK,QAAgB,QAAmD;AAC5E,SAAO,KAAK,KAAK,QAAQ,OAAO;;;;;CAMlC,MAAM,SAA2C;AAE/C,SADe,MAAM,KAAK,KAAK,UAAU,EAAE,CAAC;;;;;CAO9C,MAAM,eAA8B;AAClC,QAAM,KAAK,KAAK,aAAa,EAAE,CAAC;;;;;CAUlC,MAAM,wBAGH;AAED,SADe,MAAM,KAAK,KAAK,2BAA2B,EAAE,CAAC;;;;;CAO/D,MAAM,sBAAsB,OAIgB;AAE1C,SADe,MAAM,KAAK,KAAK,2BAA2B,MAAiC;;;;;CAO7F,MAAM,iBAAiB,SAIC;AAEtB,SADe,MAAM,KAAK,KAAK,qBAAqB,QAAmC;;;;;;CAYzF,MAAM,WAAW,QAAqD;AAEpE,SADe,MAAM,KAAK,KAAK,eAAe,OAA6C;;;;;;CAY7F,MAAM,iBAAiB,QAGa;EAIlC,MAAM,aADS,MAAM,KAAK,KAAK,sBAAsB,OAAkC;AAEvF,MAAI,WAAW,QAAS,QAAO;EAC/B,MAAM,OAAO,WAAW,UAAU,IAAI,QAAQ;AAE9C,MAAI;AACF,UAAO,KAAK,MAAM,KAAK;UACjB;AACN,UAAO;;;;;;;CAYX,AAAQ,KACN,QACA,QACkB;EAClB,MAAM,aAAa,KAAK;AAExB,SAAO,IAAI,SAAS,SAAS,WAAW;GACtC,IAAI,SAAwB;GAC5B,IAAI,OAAO;GACX,IAAI,SAAS;GACb,IAAI,QAA8C;GAElD,SAAS,OAAO,OAAqB,OAAuB;AAC1D,QAAI,KAAM;AACV,WAAO;AACP,QAAI,UAAU,MAAM;AAClB,kBAAa,MAAM;AACnB,aAAQ;;AAEV,QAAI;AACF,aAAQ,SAAS;YACX;AAGR,QAAI,MACF,QAAO,MAAM;QAEb,SAAQ,MAAM;;AAIlB,YAAS,QAAQ,kBAAkB;IACjC,MAAM,UAAsB;KAC1B,IAAI,YAAY;KAChB;KACA;KACD;AACD,WAAQ,MAAM,KAAK,UAAU,QAAQ,GAAG,KAAK;KAC7C;AAEF,UAAO,GAAG,SAAS,UAAkB;AACnC,cAAU,MAAM,UAAU;IAC1B,MAAM,KAAK,OAAO,QAAQ,KAAK;AAC/B,QAAI,OAAO,GAAI;IAEf,MAAM,OAAO,OAAO,MAAM,GAAG,GAAG;AAChC,aAAS,OAAO,MAAM,KAAK,EAAE;IAE7B,IAAI;AACJ,QAAI;AACF,gBAAW,KAAK,MAAM,KAAK;YACrB;AACN,4BAAO,IAAI,MAAM,oBAAoB,OAAO,CAAC;AAC7C;;AAGF,QAAI,CAAC,SAAS,GACZ,QAAO,IAAI,MAAM,SAAS,SAAS,kBAAkB,CAAC;QAEtD,QAAO,MAAM,SAAS,OAAO;KAE/B;AAEF,UAAO,GAAG,UAAU,MAA6B;AAC/C,QAAI,EAAE,SAAS,YAAY,EAAE,SAAS,eACpC,wBACE,IAAI,MACF,0DACD,CACF;QAED,QAAO,EAAE;KAEX;AAEF,UAAO,GAAG,aAAa;AACrB,QAAI,CAAC,KACH,wBAAO,IAAI,MAAM,wCAAwC,CAAC;KAE5D;AAEF,WAAQ,iBAAiB;AACvB,2BAAO,IAAI,MAAM,+BAA+B,CAAC;MAChD,eAAe;IAClB"}

package/dist/mcp/index.mjs

@@ -1,10 +1,10 @@
 #!/usr/bin/env node
 import { n as openRegistry } from "../db-4lSqLFb8.mjs";
-import { n as openFederation } from "../db-BcDxXVBu.mjs";
-import "../embeddings-mfqv-jFu.mjs";
-import "../search-PjftDxxs.mjs";
-import { a as record, i as number, n as array, o as string, r as boolean, s as unknown, t as _enum } from "../schemas-DjdwzIQ8.mjs";
-import { a as toolProjectHealth, d as toolSessionRoute, f as toolTopicDetect, i as toolProjectDetect, l as toolRegistrySearch, n as toolMemorySearch, o as toolProjectInfo, r as toolNotificationConfig, s as toolProjectList, t as toolMemoryGet, u as toolSessionList } from "../tools-CLK4080-.mjs";
+import { n as openFederation } from "../db-Dp8VXIMR.mjs";
+import "../embeddings-DGRAPAYb.mjs";
+import "../search-GK0ibTJy.mjs";
+import { a as record, i as number, n as array, o as string, r as boolean, s as unknown, t as _enum } from "../schemas-BY3Pjvje.mjs";
+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-CUg0Lyg-.mjs";
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
 
@@ -349,6 +349,119 @@ async function startMcpServer() {
 			isError: result.isError
 		};
 	});
+	server.tool("zettel_explore", [
+		"Explore the vault's knowledge graph using Luhmann's Folgezettel traversal.",
+		"Follow trains of thought forward, backward, or both from a starting note.",
+		"Classifies links as sequential (same-folder) or associative (cross-folder)."
+	].join("\n"), {
+		start_note: string().describe("Path or title of the note to start traversal from."),
+		depth: number().int().min(1).max(10).optional().describe("How many link hops to traverse. Default: 3."),
+		direction: _enum([
+			"forward",
+			"backward",
+			"both"
+		]).optional().describe("Traversal direction: 'forward' (outlinks), 'backward' (backlinks), or 'both'. Default: both."),
+		mode: _enum([
+			"sequential",
+			"associative",
+			"all"
+		]).optional().describe("Link type filter: 'sequential' (same-folder), 'associative' (cross-folder), or 'all'. Default: all.")
+	}, async (args) => {
+		const result = await toolZettelExplore(getFederationDb(), args);
+		return {
+			content: result.content.map((c) => ({
+				type: c.type,
+				text: c.text
+			})),
+			isError: result.isError
+		};
+	});
+	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"), {
+		scope: _enum([
+			"full",
+			"recent",
+			"project"
+		]).optional().describe("Audit scope: 'full' (entire vault), 'recent' (recently modified), or 'project' (specific path). Default: full."),
+		project_path: string().optional().describe("Absolute path to the project/folder to audit when scope='project'."),
+		recent_days: number().int().optional().describe("Number of days to look back when scope='recent'. Default: 30."),
+		include: array(_enum([
+			"dead_links",
+			"orphans",
+			"disconnected",
+			"low_connectivity"
+		])).optional().describe("Specific checks to include. Omit to run all checks.")
+	}, async (args) => {
+		const result = await toolZettelHealth(getFederationDb(), args);
+		return {
+			content: result.content.map((c) => ({
+				type: c.type,
+				text: c.text
+			})),
+			isError: result.isError
+		};
+	});
+	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"), {
+		reference_path: string().describe("Path to the reference note to find surprising connections for."),
+		vault_project_id: number().int().describe("Project ID of the vault to search within."),
+		limit: number().int().optional().describe("Maximum number of surprising notes to return. Default: 10."),
+		min_similarity: number().optional().describe("Minimum semantic similarity [0,1] for a note to be considered. Default: 0.5."),
+		min_graph_distance: number().int().optional().describe("Minimum link hops away from the reference note. Default: 3.")
+	}, async (args) => {
+		const result = await toolZettelSurprise(getFederationDb(), args);
+		return {
+			content: result.content.map((c) => ({
+				type: c.type,
+				text: c.text
+			})),
+			isError: result.isError
+		};
+	});
+	server.tool("zettel_suggest", ["Suggest new connections for a note using semantic similarity, shared tags, and graph neighborhood (friends-of-friends)."].join("\n"), {
+		note_path: string().describe("Path to the note to generate link suggestions for."),
+		vault_project_id: number().int().describe("Project ID of the vault to search within."),
+		limit: number().int().optional().describe("Maximum number of suggestions to return. Default: 10."),
+		exclude_linked: boolean().optional().describe("Exclude notes already linked from this note. Default: true.")
+	}, async (args) => {
+		const result = await toolZettelSuggest(getFederationDb(), args);
+		return {
+			content: result.content.map((c) => ({
+				type: c.type,
+				text: c.text
+			})),
+			isError: result.isError
+		};
+	});
+	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"), {
+		question: string().describe("The question or topic to explore in the vault."),
+		vault_project_id: number().int().describe("Project ID of the vault to query."),
+		depth: number().int().optional().describe("How many link hops to follow from seed notes. Default: 2."),
+		limit: number().int().optional().describe("Maximum number of relevant notes to retrieve. Default: 10.")
+	}, async (args) => {
+		const result = await toolZettelConverse(getFederationDb(), args);
+		return {
+			content: result.content.map((c) => ({
+				type: c.type,
+				text: c.text
+			})),
+			isError: result.isError
+		};
+	});
+	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"), {
+		vault_project_id: number().int().describe("Project ID of the vault to analyse."),
+		lookback_days: number().int().optional().describe("Number of days of recent notes to cluster. Default: 30."),
+		min_cluster_size: number().int().optional().describe("Minimum notes required to form a theme cluster. Default: 3."),
+		max_themes: number().int().optional().describe("Maximum number of theme clusters to return. Default: 10."),
+		similarity_threshold: number().optional().describe("Minimum cosine similarity to group notes into a cluster [0,1]. Default: 0.7.")
+	}, async (args) => {
+		const result = await toolZettelThemes(getFederationDb(), args);
+		return {
+			content: result.content.map((c) => ({
+				type: c.type,
+				text: c.text
+			})),
+			isError: result.isError
+		};
+	});
 	const transport = new StdioServerTransport();
 	await server.connect(transport);
 }