brainbank 0.2.1 → 0.3.0

package/dist/{chunk-F6SJ3U4H.js → chunk-Y3JKI6QN.js}

@@ -33,10 +33,25 @@ var BREAK_SCORES = [
 var TARGET_CHARS = 3e3;
 var WINDOW_CHARS = 600;
 var MIN_CHUNK_CHARS = 200;
-function escapeRegex(s) {
-  return s.replace(/[.+?^${}()|[\]\\]/g, "\\$&");
-}
-__name(escapeRegex, "escapeRegex");
+var IGNORED_DOC_DIRS = /* @__PURE__ */ new Set([
+  "node_modules",
+  ".git",
+  ".hg",
+  ".svn",
+  "dist",
+  "build",
+  "out",
+  "coverage",
+  ".next",
+  "__pycache__",
+  ".tox",
+  ".venv",
+  "venv",
+  "vendor",
+  "target",
+  ".cache",
+  ".turbo"
+]);
 var DocsIndexer = class {
   constructor(_db, _embedding, _hnsw, _vecCache) {
     this._db = _db;
@@ -56,9 +71,30 @@ var DocsIndexer = class {
     if (!fs.existsSync(absDir)) {
       throw new Error(`Collection path does not exist: ${absDir}`);
     }
+    const files = this._walkFiles(absDir, pattern, options.ignore);
+    let indexed = 0, skipped = 0, totalChunks = 0;
+    for (let i = 0; i < files.length; i++) {
+      const relPath = files[i];
+      options.onProgress?.(relPath, i + 1, files.length);
+      const absPath = path.join(absDir, relPath);
+      const content = fs.readFileSync(absPath, "utf-8");
+      const hash = createHash("sha256").update(content).digest("hex").slice(0, 16);
+      if (this._isUnchanged(collection, relPath, hash)) {
+        skipped++;
+        continue;
+      }
+      this._removeOldChunks(collection, relPath);
+      const chunkCount = await this._indexFile(collection, relPath, content, hash);
+      indexed++;
+      totalChunks += chunkCount;
+    }
+    return { indexed, skipped, chunks: totalChunks };
+  }
+  /** Walk directory tree and collect matching files. */
+  _walkFiles(absDir, pattern, ignore) {
     const patternExt = pattern.match(/\.([\w]+)$/)?.[1];
     const files = [];
-    const walkDir = /* @__PURE__ */ __name((dir, base) => {
+    const walk = /* @__PURE__ */ __name((dir, base) => {
       let entries;
       try {
         entries = fs.readdirSync(dir, { withFileTypes: true });
@@ -68,93 +104,89 @@ var DocsIndexer = class {
       for (const e of entries) {
         const rel = base ? `${base}/${e.name}` : e.name;
         if (e.isDirectory()) {
-          if (this._isIgnoredDocDir(e.name)) continue;
-          walkDir(path.join(dir, e.name), rel);
+          if (IGNORED_DOC_DIRS.has(e.name)) continue;
+          walk(path.join(dir, e.name), rel);
         } else if (e.isFile()) {
-          const shouldIgnore = options.ignore?.some((ig) => {
-            const escaped = escapeRegex(ig).replace(/\\\*\\\*/g, ".*").replace(/\\\*/g, "[^/]*");
-            return new RegExp(escaped).test(rel);
-          });
+          if (this._isIgnoredFile(rel, ignore)) continue;
           const ext = path.extname(e.name).slice(1);
-          if (!shouldIgnore && (!patternExt || ext === patternExt)) {
-            files.push(rel);
-          }
+          if (!patternExt || ext === patternExt) files.push(rel);
         }
       }
-    }, "walkDir");
-    walkDir(absDir, "");
-    let indexed = 0;
-    let skipped = 0;
-    let totalChunks = 0;
-    for (let i = 0; i < files.length; i++) {
-      const relPath = files[i];
-      const absPath = path.join(absDir, relPath);
-      options.onProgress?.(relPath, i + 1, files.length);
-      const content = fs.readFileSync(absPath, "utf-8");
-      const hash = createHash("sha256").update(content).digest("hex").slice(0, 16);
-      const existingChunks = this._db.prepare(
-        `SELECT dc.id, dc.content_hash, dv.chunk_id AS has_vector
-                 FROM doc_chunks dc
-                 LEFT JOIN doc_vectors dv ON dv.chunk_id = dc.id
-                 WHERE dc.collection = ? AND dc.file_path = ?`
-      ).all(collection, relPath);
-      const allMatch = existingChunks.length > 0 && existingChunks.every((c) => c.content_hash === hash && c.has_vector != null);
-      if (allMatch) {
-        skipped++;
-        continue;
-      }
-      for (const old of existingChunks) {
-        this._hnsw.remove(old.id);
-        this._vecCache.delete(old.id);
+    }, "walk");
+    walk(absDir, "");
+    return files;
+  }
+  /** Check if a file matches any ignore patterns. */
+  _isIgnoredFile(relPath, ignore) {
+    if (!ignore) return false;
+    return ignore.some((ig) => {
+      const regex = ig.replace(/\*\*/g, "{{GLOBSTAR}}").replace(/\*/g, "{{STAR}}").replace(/[.+?^${}()|[\]\\]/g, "\\$&").replace(/\{\{GLOBSTAR\}\}/g, ".*").replace(/\{\{STAR\}\}/g, "[^/]*");
+      return new RegExp(regex).test(relPath);
+    });
+  }
+  /** Check if all chunks for a file match the current hash and have vectors. */
+  _isUnchanged(collection, relPath, hash) {
+    const existing = this._db.prepare(
+      `SELECT dc.id, dc.content_hash, dv.chunk_id AS has_vector
+             FROM doc_chunks dc
+             LEFT JOIN doc_vectors dv ON dv.chunk_id = dc.id
+             WHERE dc.collection = ? AND dc.file_path = ?`
+    ).all(collection, relPath);
+    return existing.length > 0 && existing.every((c) => c.content_hash === hash && c.has_vector != null);
+  }
+  /** Remove old chunks and their HNSW vectors for a file. */
+  _removeOldChunks(collection, relPath) {
+    const oldChunks = this._db.prepare(
+      "SELECT id FROM doc_chunks WHERE collection = ? AND file_path = ?"
+    ).all(collection, relPath);
+    for (const old of oldChunks) {
+      this._hnsw.remove(old.id);
+      this._vecCache.delete(old.id);
+    }
+    this._db.prepare(
+      "DELETE FROM doc_chunks WHERE collection = ? AND file_path = ?"
+    ).run(collection, relPath);
+  }
+  /** Index a single file: chunk, embed, store in DB + HNSW. */
+  async _indexFile(collection, relPath, content, hash) {
+    const title = this._extractTitle(content, relPath);
+    const chunks = this._smartChunk(content);
+    const insertChunk = this._db.prepare(`
+            INSERT INTO doc_chunks (collection, file_path, title, content, seq, pos, content_hash)
+            VALUES (?, ?, ?, ?, ?, ?, ?)
+        `);
+    const chunkIds = [];
+    this._db.transaction(() => {
+      for (let seq = 0; seq < chunks.length; seq++) {
+        const result = insertChunk.run(
+          collection,
+          relPath,
+          title,
+          chunks[seq].text,
+          seq,
+          chunks[seq].pos,
+          hash
+        );
+        chunkIds.push(Number(result.lastInsertRowid));
       }
-      this._db.prepare(
-        "DELETE FROM doc_chunks WHERE collection = ? AND file_path = ?"
-      ).run(collection, relPath);
-      const title = this._extractTitle(content, relPath);
-      const chunks = this._smartChunk(content);
-      const insertChunk = this._db.prepare(`
-                INSERT INTO doc_chunks (collection, file_path, title, content, seq, pos, content_hash)
-                VALUES (?, ?, ?, ?, ?, ?, ?)
-            `);
-      const chunkIds = [];
-      this._db.transaction(() => {
-        for (let seq = 0; seq < chunks.length; seq++) {
-          const chunk = chunks[seq];
-          const result = insertChunk.run(
-            collection,
-            relPath,
-            title,
-            chunk.text,
-            seq,
-            chunk.pos,
-            hash
-          );
-          chunkIds.push(Number(result.lastInsertRowid));
-        }
-      });
-      const texts = chunks.map((c) => `title: ${title} | text: ${c.text}`);
-      const embeddings = await this._embedding.embedBatch(texts);
-      const insertVec = this._db.prepare(
-        "INSERT OR REPLACE INTO doc_vectors (chunk_id, embedding) VALUES (?, ?)"
-      );
-      this._db.transaction(() => {
-        for (let j = 0; j < chunkIds.length; j++) {
-          const buf = Buffer.from(embeddings[j].buffer);
-          insertVec.run(chunkIds[j], buf);
-        }
-      });
+    });
+    const texts = chunks.map((c) => `title: ${title} | text: ${c.text}`);
+    const embeddings = await this._embedding.embedBatch(texts);
+    const insertVec = this._db.prepare(
+      "INSERT OR REPLACE INTO doc_vectors (chunk_id, embedding) VALUES (?, ?)"
+    );
+    this._db.transaction(() => {
       for (let j = 0; j < chunkIds.length; j++) {
-        this._hnsw.add(embeddings[j], chunkIds[j]);
-        this._vecCache.set(chunkIds[j], embeddings[j]);
+        insertVec.run(chunkIds[j], Buffer.from(embeddings[j].buffer));
       }
-      indexed++;
-      totalChunks += chunks.length;
+    });
+    for (let j = 0; j < chunkIds.length; j++) {
+      this._hnsw.add(embeddings[j], chunkIds[j]);
+      this._vecCache.set(chunkIds[j], embeddings[j]);
     }
-    return { indexed, skipped, chunks: totalChunks };
+    return chunks.length;
   }
-  /**
-   * Remove all indexed data for a collection.
-   */
+  /** Remove all indexed data for a collection. */
   removeCollection(collection) {
     const chunks = this._db.prepare(
       "SELECT id FROM doc_chunks WHERE collection = ?"
@@ -168,10 +200,7 @@ var DocsIndexer = class {
     this._db.prepare("DELETE FROM path_contexts WHERE collection = ?").run(collection);
   }
   // ── Smart Chunking ──────────────────────────────
-  /**
-   * Split document into chunks at natural markdown boundaries.
-   * Uses heading-aware scoring like qmd.
-   */
+  /** Split document into chunks at natural markdown boundaries. */
   _smartChunk(text) {
     if (text.length <= TARGET_CHARS) {
       return [{ text, pos: 0 }];
@@ -183,32 +212,10 @@ var DocsIndexer = class {
     while (chunkStart < text.length) {
       const remaining = text.length - chunkStart;
       if (remaining <= TARGET_CHARS + WINDOW_CHARS) {
-        const lastText = text.slice(chunkStart).trim();
-        if (lastText.length >= MIN_CHUNK_CHARS) {
-          chunks.push({ text: lastText, pos: chunkStart });
-        } else if (chunks.length > 0) {
-          chunks[chunks.length - 1].text += "\n" + lastText;
-        } else {
-          chunks.push({ text: lastText, pos: chunkStart });
-        }
+        this._pushLastChunk(text, chunkStart, chunks);
         break;
       }
-      const targetEnd = chunkStart + TARGET_CHARS;
-      const windowStart = targetEnd - WINDOW_CHARS;
-      let bestBreak = targetEnd;
-      let bestScore = 0;
-      for (const bp of breakPoints) {
-        if (bp.pos <= chunkStart) continue;
-        if (bp.pos > targetEnd + WINDOW_CHARS / 2) break;
-        if (bp.pos < windowStart) continue;
-        const distance = Math.abs(bp.pos - targetEnd);
-        const decay = 1 - (distance / WINDOW_CHARS) ** 2 * 0.7;
-        const finalScore = bp.score * decay;
-        if (finalScore > bestScore) {
-          bestScore = finalScore;
-          bestBreak = bp.pos;
-        }
-      }
+      const bestBreak = this._findBestBreak(chunkStart, breakPoints);
       const chunkText = text.slice(chunkStart, bestBreak).trim();
       if (chunkText.length >= MIN_CHUNK_CHARS) {
         chunks.push({ text: chunkText, pos: chunkStart });
@@ -217,9 +224,38 @@ var DocsIndexer = class {
     }
     return chunks;
   }
-  /**
-   * Find all potential break points in the document with scores.
-   */
+  /** Handle the last chunk: merge if too small, otherwise push. */
+  _pushLastChunk(text, chunkStart, chunks) {
+    const lastText = text.slice(chunkStart).trim();
+    if (lastText.length >= MIN_CHUNK_CHARS) {
+      chunks.push({ text: lastText, pos: chunkStart });
+    } else if (chunks.length > 0) {
+      chunks[chunks.length - 1].text += "\n" + lastText;
+    } else {
+      chunks.push({ text: lastText, pos: chunkStart });
+    }
+  }
+  /** Find the best break position within the target window. */
+  _findBestBreak(chunkStart, breakPoints) {
+    const targetEnd = chunkStart + TARGET_CHARS;
+    const windowStart = targetEnd - WINDOW_CHARS;
+    let bestBreak = targetEnd;
+    let bestScore = 0;
+    for (const bp of breakPoints) {
+      if (bp.pos <= chunkStart) continue;
+      if (bp.pos > targetEnd + WINDOW_CHARS / 2) break;
+      if (bp.pos < windowStart) continue;
+      const distance = Math.abs(bp.pos - targetEnd);
+      const decay = 1 - (distance / WINDOW_CHARS) ** 2 * 0.7;
+      const finalScore = bp.score * decay;
+      if (finalScore > bestScore) {
+        bestScore = finalScore;
+        bestBreak = bp.pos;
+      }
+    }
+    return bestBreak;
+  }
+  /** Find all potential break points in the document with scores. */
   _findBreakPoints(lines) {
     const points = [];
     let charPos = 0;
@@ -247,37 +283,12 @@ var DocsIndexer = class {
     }
     return points;
   }
-  /**
-   * Extract document title from first heading or filename.
-   */
+  /** Extract document title from first heading or filename. */
   _extractTitle(content, filePath) {
     const match = content.match(/^#{1,3}\s+(.+)$/m);
     if (match) return match[1].trim();
     return path.basename(filePath, path.extname(filePath));
   }
-  /** Skip well-known output/vendor directories when walking docs. */
-  _isIgnoredDocDir(name) {
-    const IGNORED = /* @__PURE__ */ new Set([
-      "node_modules",
-      ".git",
-      ".hg",
-      ".svn",
-      "dist",
-      "build",
-      "out",
-      "coverage",
-      ".next",
-      "__pycache__",
-      ".tox",
-      ".venv",
-      "venv",
-      "vendor",
-      "target",
-      ".cache",
-      ".turbo"
-    ]);
-    return IGNORED.has(name);
-  }
 };
 
 // src/indexers/docs/docs-plugin.ts
@@ -434,4 +445,4 @@ export {
   DocsIndexer,
   docs
 };
-//# sourceMappingURL=chunk-F6SJ3U4H.js.map
+//# sourceMappingURL=chunk-Y3JKI6QN.js.map

package/dist/chunk-Y3JKI6QN.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/indexers/docs/docs-indexer.ts","../src/indexers/docs/docs-plugin.ts"],"sourcesContent":["/**\n * BrainBank — Document Indexer\n * \n * Indexes generic document collections (markdown, text, etc.)\n * with heading-aware smart chunking, inspired by qmd.\n * \n *   const indexer = new DocsIndexer(db, embedding, hnsw, vecCache);\n *   await indexer.indexCollection('notes', '/path/to/notes', '**\\/*.md');\n */\n\nimport * as fs from 'node:fs';\nimport * as path from 'node:path';\nimport { createHash } from 'node:crypto';\n\nimport type { Database } from '@/db/database.ts';\nimport type { EmbeddingProvider, VectorIndex } from '@/types.ts';\nimport type { HNSWIndex } from '@/providers/vector/hnsw-index.ts';\n\n// ── Break Point Scoring (qmd-inspired) ──────────────\n\ninterface BreakPoint {\n    pos: number;      // character position\n    score: number;    // break quality (higher = better)\n}\n\nconst BREAK_SCORES: [RegExp, number][] = [\n    [/^# /,      100],   // H1\n    [/^## /,      90],   // H2\n    [/^### /,     80],   // H3\n    [/^#### /,    70],   // H4\n    [/^##### /,   60],   // H5\n    [/^###### /,  50],   // H6\n    [/^```/,      80],   // Code fence\n    [/^---$/,     60],   // Horizontal rule\n    [/^\\*\\*\\*$/,  60],   // Horizontal rule alt\n    [/^$/,        20],   // Blank line (paragraph break)\n    [/^[-*+] /,    5],   // List item\n];\n\n// ── Chunk Target ────────────────────────────────────\n\nconst TARGET_CHARS = 3000;       // ~900 tokens\nconst WINDOW_CHARS = 600;        // search window before cutoff\nconst MIN_CHUNK_CHARS = 200;     // don't create tiny chunks\n\n/** Escape special regex characters so user-provided patterns behave as literals. */\nfunction escapeRegex(s: string): string {\n    return s.replace(/[.+?^${}()|[\\]\\\\]/g, '\\\\$&');\n}\n\n/** Ignored output/vendor directories when walking docs. */\nconst IGNORED_DOC_DIRS = new Set([\n    'node_modules', '.git', '.hg', '.svn',\n    'dist', 'build', 'out', 'coverage', '.next',\n    '__pycache__', '.tox', '.venv', 'venv',\n    'vendor', 'target', '.cache', '.turbo',\n]);\n\n// ── DocsIndexer ──────────────────────────────────────\n\nexport class DocsIndexer {\n    constructor(\n        private _db: Database,\n        private _embedding: EmbeddingProvider,\n        private _hnsw: HNSWIndex,\n        private _vecCache: Map<number, Float32Array>,\n    ) {}\n\n    /**\n     * Index all documents in a collection.\n     * Incremental — skips unchanged files (by content hash).\n     */\n    async indexCollection(\n        collection: string,\n        dirPath: string,\n        pattern: string = '**/*.md',\n        options: {\n            ignore?: string[];\n            onProgress?: (file: string, current: number, total: number) => void;\n        } = {},\n    ): Promise<{ indexed: number; skipped: number; chunks: number }> {\n        const absDir = path.resolve(dirPath);\n        if (!fs.existsSync(absDir)) {\n            throw new Error(`Collection path does not exist: ${absDir}`);\n        }\n\n        const files = this._walkFiles(absDir, pattern, options.ignore);\n        let indexed = 0, skipped = 0, totalChunks = 0;\n\n        for (let i = 0; i < files.length; i++) {\n            const relPath = files[i];\n            options.onProgress?.(relPath, i + 1, files.length);\n\n            const absPath = path.join(absDir, relPath);\n            const content = fs.readFileSync(absPath, 'utf-8');\n            const hash = createHash('sha256').update(content).digest('hex').slice(0, 16);\n\n            if (this._isUnchanged(collection, relPath, hash)) {\n                skipped++;\n                continue;\n            }\n\n            this._removeOldChunks(collection, relPath);\n            const chunkCount = await this._indexFile(collection, relPath, content, hash);\n            indexed++;\n            totalChunks += chunkCount;\n        }\n\n        return { indexed, skipped, chunks: totalChunks };\n    }\n\n    /** Walk directory tree and collect matching files. */\n    private _walkFiles(absDir: string, pattern: string, ignore?: string[]): string[] {\n        const patternExt = pattern.match(/\\.([\\w]+)$/)?.[1];\n        const files: string[] = [];\n\n        const walk = (dir: string, base: string): void => {\n            let entries: fs.Dirent[];\n            try { entries = fs.readdirSync(dir, { withFileTypes: true }); }\n            catch { return; }\n            for (const e of entries) {\n                const rel = base ? `${base}/${e.name}` : e.name;\n                if (e.isDirectory()) {\n                    if (IGNORED_DOC_DIRS.has(e.name)) continue;\n                    walk(path.join(dir, e.name), rel);\n                } else if (e.isFile()) {\n                    if (this._isIgnoredFile(rel, ignore)) continue;\n                    const ext = path.extname(e.name).slice(1);\n                    if (!patternExt || ext === patternExt) files.push(rel);\n                }\n            }\n        };\n        walk(absDir, '');\n        return files;\n    }\n\n    /** Check if a file matches any ignore patterns. */\n    private _isIgnoredFile(relPath: string, ignore?: string[]): boolean {\n        if (!ignore) return false;\n        return ignore.some(ig => {\n            const regex = ig\n                .replace(/\\*\\*/g, '{{GLOBSTAR}}')\n                .replace(/\\*/g, '{{STAR}}')\n                .replace(/[.+?^${}()|[\\]\\\\]/g, '\\\\$&')\n                .replace(/\\{\\{GLOBSTAR\\}\\}/g, '.*')\n                .replace(/\\{\\{STAR\\}\\}/g, '[^/]*');\n            return new RegExp(regex).test(relPath);\n        });\n    }\n\n    /** Check if all chunks for a file match the current hash and have vectors. */\n    private _isUnchanged(collection: string, relPath: string, hash: string): boolean {\n        const existing = this._db.prepare(\n            `SELECT dc.id, dc.content_hash, dv.chunk_id AS has_vector\n             FROM doc_chunks dc\n             LEFT JOIN doc_vectors dv ON dv.chunk_id = dc.id\n             WHERE dc.collection = ? AND dc.file_path = ?`\n        ).all(collection, relPath) as any[];\n\n        return existing.length > 0 &&\n            existing.every((c: any) => c.content_hash === hash && c.has_vector != null);\n    }\n\n    /** Remove old chunks and their HNSW vectors for a file. */\n    private _removeOldChunks(collection: string, relPath: string): void {\n        const oldChunks = this._db.prepare(\n            'SELECT id FROM doc_chunks WHERE collection = ? AND file_path = ?'\n        ).all(collection, relPath) as any[];\n\n        for (const old of oldChunks) {\n            this._hnsw.remove(old.id);\n            this._vecCache.delete(old.id);\n        }\n        this._db.prepare(\n            'DELETE FROM doc_chunks WHERE collection = ? AND file_path = ?'\n        ).run(collection, relPath);\n    }\n\n    /** Index a single file: chunk, embed, store in DB + HNSW. */\n    private async _indexFile(\n        collection: string, relPath: string, content: string, hash: string,\n    ): Promise<number> {\n        const title = this._extractTitle(content, relPath);\n        const chunks = this._smartChunk(content);\n\n        const insertChunk = this._db.prepare(`\n            INSERT INTO doc_chunks (collection, file_path, title, content, seq, pos, content_hash)\n            VALUES (?, ?, ?, ?, ?, ?, ?)\n        `);\n\n        const chunkIds: number[] = [];\n        this._db.transaction(() => {\n            for (let seq = 0; seq < chunks.length; seq++) {\n                const result = insertChunk.run(\n                    collection, relPath, title, chunks[seq].text, seq, chunks[seq].pos, hash,\n                );\n                chunkIds.push(Number(result.lastInsertRowid));\n            }\n        });\n\n        const texts = chunks.map(c => `title: ${title} | text: ${c.text}`);\n        const embeddings = await this._embedding.embedBatch(texts);\n\n        const insertVec = this._db.prepare(\n            'INSERT OR REPLACE INTO doc_vectors (chunk_id, embedding) VALUES (?, ?)'\n        );\n        this._db.transaction(() => {\n            for (let j = 0; j < chunkIds.length; j++) {\n                insertVec.run(chunkIds[j], Buffer.from(embeddings[j].buffer));\n            }\n        });\n\n        for (let j = 0; j < chunkIds.length; j++) {\n            this._hnsw.add(embeddings[j], chunkIds[j]);\n            this._vecCache.set(chunkIds[j], embeddings[j]);\n        }\n\n        return chunks.length;\n    }\n\n    /** Remove all indexed data for a collection. */\n    removeCollection(collection: string): void {\n        const chunks = this._db.prepare(\n            'SELECT id FROM doc_chunks WHERE collection = ?'\n        ).all(collection) as any[];\n        for (const chunk of chunks) {\n            this._hnsw.remove(chunk.id);\n            this._vecCache.delete(chunk.id);\n        }\n\n        this._db.prepare('DELETE FROM doc_chunks WHERE collection = ?').run(collection);\n        this._db.prepare('DELETE FROM collections WHERE name = ?').run(collection);\n        this._db.prepare('DELETE FROM path_contexts WHERE collection = ?').run(collection);\n    }\n\n    // ── Smart Chunking ──────────────────────────────\n\n    /** Split document into chunks at natural markdown boundaries. */\n    private _smartChunk(text: string): { text: string; pos: number }[] {\n        if (text.length <= TARGET_CHARS) {\n            return [{ text, pos: 0 }];\n        }\n\n        const lines = text.split('\\n');\n        const breakPoints = this._findBreakPoints(lines);\n        const chunks: { text: string; pos: number }[] = [];\n        let chunkStart = 0;\n\n        while (chunkStart < text.length) {\n            const remaining = text.length - chunkStart;\n            if (remaining <= TARGET_CHARS + WINDOW_CHARS) {\n                this._pushLastChunk(text, chunkStart, chunks);\n                break;\n            }\n\n            const bestBreak = this._findBestBreak(chunkStart, breakPoints);\n            const chunkText = text.slice(chunkStart, bestBreak).trim();\n            if (chunkText.length >= MIN_CHUNK_CHARS) {\n                chunks.push({ text: chunkText, pos: chunkStart });\n            }\n            chunkStart = bestBreak;\n        }\n\n        return chunks;\n    }\n\n    /** Handle the last chunk: merge if too small, otherwise push. */\n    private _pushLastChunk(\n        text: string, chunkStart: number, chunks: { text: string; pos: number }[],\n    ): void {\n        const lastText = text.slice(chunkStart).trim();\n        if (lastText.length >= MIN_CHUNK_CHARS) {\n            chunks.push({ text: lastText, pos: chunkStart });\n        } else if (chunks.length > 0) {\n            chunks[chunks.length - 1].text += '\\n' + lastText;\n        } else {\n            chunks.push({ text: lastText, pos: chunkStart });\n        }\n    }\n\n    /** Find the best break position within the target window. */\n    private _findBestBreak(chunkStart: number, breakPoints: BreakPoint[]): number {\n        const targetEnd = chunkStart + TARGET_CHARS;\n        const windowStart = targetEnd - WINDOW_CHARS;\n\n        let bestBreak = targetEnd;\n        let bestScore = 0;\n\n        for (const bp of breakPoints) {\n            if (bp.pos <= chunkStart) continue;\n            if (bp.pos > targetEnd + WINDOW_CHARS / 2) break;\n            if (bp.pos < windowStart) continue;\n\n            const distance = Math.abs(bp.pos - targetEnd);\n            const decay = 1 - (distance / WINDOW_CHARS) ** 2 * 0.7;\n            const finalScore = bp.score * decay;\n\n            if (finalScore > bestScore) {\n                bestScore = finalScore;\n                bestBreak = bp.pos;\n            }\n        }\n\n        return bestBreak;\n    }\n\n    /** Find all potential break points in the document with scores. */\n    private _findBreakPoints(lines: string[]): BreakPoint[] {\n        const points: BreakPoint[] = [];\n        let charPos = 0;\n        let inCodeBlock = false;\n\n        for (const line of lines) {\n            if (line.trimStart().startsWith('```')) {\n                inCodeBlock = !inCodeBlock;\n                if (!inCodeBlock) {\n                    points.push({ pos: charPos + line.length + 1, score: 80 });\n                }\n                charPos += line.length + 1;\n                continue;\n            }\n\n            if (inCodeBlock) {\n                charPos += line.length + 1;\n                continue;\n            }\n\n            for (const [pattern, score] of BREAK_SCORES) {\n                if (pattern.test(line.trim())) {\n                    points.push({ pos: charPos, score });\n                    break;\n                }\n            }\n\n            charPos += line.length + 1;\n        }\n\n        return points;\n    }\n\n    /** Extract document title from first heading or filename. */\n    private _extractTitle(content: string, filePath: string): string {\n        const match = content.match(/^#{1,3}\\s+(.+)$/m);\n        if (match) return match[1].trim();\n        return path.basename(filePath, path.extname(filePath));\n    }\n}\n","/**\n * BrainBank — Docs Module\n * \n * Index any folder of markdown/text files (notes, docs, wikis).\n * Heading-aware smart chunking inspired by qmd.\n * \n *   import { docs } from 'brainbank/docs';\n *   brain.use(docs());\n */\n\nimport type { Indexer, IndexerContext } from '@/indexers/base.ts';\nimport type { HNSWIndex } from '@/providers/vector/hnsw-index.ts';\nimport type { Database } from '@/db/database.ts';\nimport type { EmbeddingProvider, DocumentCollection, SearchResult } from '@/types.ts';\nimport { DocsIndexer } from './docs-indexer.ts';\n\nclass DocsPlugin implements Indexer {\n    readonly name = 'docs';\n    hnsw!: HNSWIndex;\n    indexer!: DocsIndexer;\n    vecCache = new Map<number, Float32Array>();\n    private _db!: Database;\n    private _embedding!: EmbeddingProvider;\n\n    async initialize(ctx: IndexerContext): Promise<void> {\n        this._db = ctx.db;\n        this._embedding = ctx.embedding;\n        this.hnsw = await ctx.createHnsw();\n        ctx.loadVectors('doc_vectors', 'chunk_id', this.hnsw, this.vecCache);\n        this.indexer = new DocsIndexer(ctx.db, ctx.embedding, this.hnsw, this.vecCache);\n    }\n\n    /** Register a document collection. */\n    addCollection(collection: DocumentCollection): void {\n        this._db.prepare(`\n            INSERT OR REPLACE INTO collections (name, path, pattern, ignore_json, context)\n            VALUES (?, ?, ?, ?, ?)\n        `).run(\n            collection.name,\n            collection.path,\n            collection.pattern ?? '**/*.md',\n            JSON.stringify(collection.ignore ?? []),\n            collection.context ?? null,\n        );\n    }\n\n    /** Remove a collection and its indexed data. */\n    removeCollection(name: string): void {\n        this.indexer.removeCollection(name);\n    }\n\n    /** List all registered collections. */\n    listCollections(): DocumentCollection[] {\n        return (this._db.prepare('SELECT * FROM collections').all() as any[]).map(row => ({\n            name: row.name,\n            path: row.path,\n            pattern: row.pattern,\n            ignore: JSON.parse(row.ignore_json),\n            context: row.context,\n        }));\n    }\n\n    /** Index all (or specific) collections. Incremental. */\n    async indexCollections(options: {\n        collections?: string[];\n        onProgress?: (collection: string, file: string, current: number, total: number) => void;\n    } = {}): Promise<Record<string, { indexed: number; skipped: number; chunks: number }>> {\n        const allCollections = this.listCollections();\n        const toIndex = options.collections\n            ? allCollections.filter(c => options.collections!.includes(c.name))\n            : allCollections;\n\n        const results: Record<string, { indexed: number; skipped: number; chunks: number }> = {};\n\n        for (const coll of toIndex) {\n            results[coll.name] = await this.indexer.indexCollection(\n                coll.name,\n                coll.path,\n                coll.pattern,\n                {\n                    ignore: coll.ignore,\n                    onProgress: (file, cur, total) => options.onProgress?.(coll.name, file, cur, total),\n                },\n            );\n        }\n\n        return results;\n    }\n\n    /** Search documents only. */\n    async search(query: string, options?: {\n        collection?: string;\n        k?: number;\n        minScore?: number;\n    }): Promise<SearchResult[]> {\n        const k = options?.k ?? 8;\n        const queryVec = await this._embedding.embed(query);\n\n        // Over-fetch from shared HNSW when filtering by collection\n        // (same pattern as collection.ts ratio scaling)\n        let searchK = k;\n        if (options?.collection && this.hnsw.size > 0) {\n            const collectionCount = (this._db.prepare(\n                'SELECT COUNT(*) as c FROM doc_chunks WHERE collection = ?'\n            ).get(options.collection) as any)?.c ?? 0;\n            const totalChunks = (this._db.prepare(\n                'SELECT COUNT(*) as c FROM doc_chunks'\n            ).get() as any)?.c ?? 1;\n            const ratio = collectionCount > 0\n                ? Math.max(3, Math.min(50, Math.ceil(totalChunks / collectionCount)))\n                : 3;\n            searchK = Math.min(k * ratio, this.hnsw.size);\n        }\n\n        const hits = this.hnsw.search(queryVec, searchK);\n\n        const results: SearchResult[] = [];\n        for (const hit of hits) {\n            if (options?.minScore && hit.score < options.minScore) continue;\n\n            const chunk = this._db.prepare(\n                'SELECT * FROM doc_chunks WHERE id = ?'\n            ).get(hit.id) as any;\n\n            if (!chunk) continue;\n            if (options?.collection && chunk.collection !== options.collection) continue;\n\n            const ctx = this._getDocContext(chunk.collection, chunk.file_path);\n\n            results.push({\n                type: 'document',\n                score: hit.score,\n                filePath: chunk.file_path,\n                content: chunk.content,\n                context: ctx,\n                metadata: {\n                    collection: chunk.collection,\n                    title: chunk.title,\n                    seq: chunk.seq,\n                },\n            });\n\n            // Stop once we have enough results\n            if (results.length >= k) break;\n        }\n\n        return results;\n    }\n\n    /** Add context description for a document path. */\n    addContext(collection: string, path: string, context: string): void {\n        this._db.prepare(`\n            INSERT OR REPLACE INTO path_contexts (collection, path, context)\n            VALUES (?, ?, ?)\n        `).run(collection, path, context);\n    }\n\n    /** Remove context for a path. */\n    removeContext(collection: string, path: string): void {\n        this._db.prepare(\n            'DELETE FROM path_contexts WHERE collection = ? AND path = ?'\n        ).run(collection, path);\n    }\n\n    /** List all context entries. */\n    listContexts(): { collection: string; path: string; context: string }[] {\n        return this._db.prepare('SELECT * FROM path_contexts').all() as any[];\n    }\n\n    stats(): Record<string, any> {\n        return {\n            collections: (this._db.prepare('SELECT COUNT(*) as c FROM collections').get() as any).c,\n            documents: (this._db.prepare('SELECT COUNT(DISTINCT file_path) as c FROM doc_chunks').get() as any).c,\n            chunks: (this._db.prepare('SELECT COUNT(*) as c FROM doc_chunks').get() as any).c,\n            hnswSize: this.hnsw.size,\n        };\n    }\n\n    /** Resolve context for a document (checks path_contexts tree → collection context). */\n    private _getDocContext(collection: string, filePath: string): string | undefined {\n        const parts = filePath.split('/');\n        for (let i = parts.length; i >= 0; i--) {\n            const checkPath = i === 0 ? '/' : '/' + parts.slice(0, i).join('/');\n            const ctx = this._db.prepare(\n                'SELECT context FROM path_contexts WHERE collection = ? AND path = ?'\n            ).get(collection, checkPath) as any;\n            if (ctx) return ctx.context;\n        }\n\n        const coll = this._db.prepare(\n            'SELECT context FROM collections WHERE name = ?'\n        ).get(collection) as any;\n        return coll?.context ?? undefined;\n    }\n}\n\n/** Create a document collections plugin. */\nexport function docs(): Indexer {\n    return new DocsPlugin();\n}\n"],"mappings":";;;;;AAUA,YAAY,QAAQ;AACpB,YAAY,UAAU;AACtB,SAAS,kBAAkB;AAa3B,IAAM,eAAmC;AAAA,EACrC,CAAC,OAAY,GAAG;AAAA;AAAA,EAChB,CAAC,QAAa,EAAE;AAAA;AAAA,EAChB,CAAC,SAAa,EAAE;AAAA;AAAA,EAChB,CAAC,UAAa,EAAE;AAAA;AAAA,EAChB,CAAC,WAAa,EAAE;AAAA;AAAA,EAChB,CAAC,YAAa,EAAE;AAAA;AAAA,EAChB,CAAC,QAAa,EAAE;AAAA;AAAA,EAChB,CAAC,SAAa,EAAE;AAAA;AAAA,EAChB,CAAC,YAAa,EAAE;AAAA;AAAA,EAChB,CAAC,MAAa,EAAE;AAAA;AAAA,EAChB,CAAC,WAAc,CAAC;AAAA;AACpB;AAIA,IAAM,eAAe;AACrB,IAAM,eAAe;AACrB,IAAM,kBAAkB;AAQxB,IAAM,mBAAmB,oBAAI,IAAI;AAAA,EAC7B;AAAA,EAAgB;AAAA,EAAQ;AAAA,EAAO;AAAA,EAC/B;AAAA,EAAQ;AAAA,EAAS;AAAA,EAAO;AAAA,EAAY;AAAA,EACpC;AAAA,EAAe;AAAA,EAAQ;AAAA,EAAS;AAAA,EAChC;AAAA,EAAU;AAAA,EAAU;AAAA,EAAU;AAClC,CAAC;AAIM,IAAM,cAAN,MAAkB;AAAA,EACrB,YACY,KACA,YACA,OACA,WACV;AAJU;AACA;AACA;AACA;AAAA,EACT;AAAA,EAlEP,OA4DyB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAYrB,MAAM,gBACF,YACA,SACA,UAAkB,WAClB,UAGI,CAAC,GACwD;AAC7D,UAAM,SAAc,aAAQ,OAAO;AACnC,QAAI,CAAI,cAAW,MAAM,GAAG;AACxB,YAAM,IAAI,MAAM,mCAAmC,MAAM,EAAE;AAAA,IAC/D;AAEA,UAAM,QAAQ,KAAK,WAAW,QAAQ,SAAS,QAAQ,MAAM;AAC7D,QAAI,UAAU,GAAG,UAAU,GAAG,cAAc;AAE5C,aAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACnC,YAAM,UAAU,MAAM,CAAC;AACvB,cAAQ,aAAa,SAAS,IAAI,GAAG,MAAM,MAAM;AAEjD,YAAM,UAAe,UAAK,QAAQ,OAAO;AACzC,YAAM,UAAa,gBAAa,SAAS,OAAO;AAChD,YAAM,OAAO,WAAW,QAAQ,EAAE,OAAO,OAAO,EAAE,OAAO,KAAK,EAAE,MAAM,GAAG,EAAE;AAE3E,UAAI,KAAK,aAAa,YAAY,SAAS,IAAI,GAAG;AAC9C;AACA;AAAA,MACJ;AAEA,WAAK,iBAAiB,YAAY,OAAO;AACzC,YAAM,aAAa,MAAM,KAAK,WAAW,YAAY,SAAS,SAAS,IAAI;AAC3E;AACA,qBAAe;AAAA,IACnB;AAEA,WAAO,EAAE,SAAS,SAAS,QAAQ,YAAY;AAAA,EACnD;AAAA;AAAA,EAGQ,WAAW,QAAgB,SAAiB,QAA6B;AAC7E,UAAM,aAAa,QAAQ,MAAM,YAAY,IAAI,CAAC;AAClD,UAAM,QAAkB,CAAC;AAEzB,UAAM,OAAO,wBAAC,KAAa,SAAuB;AAC9C,UAAI;AACJ,UAAI;AAAE,kBAAa,eAAY,KAAK,EAAE,eAAe,KAAK,CAAC;AAAA,MAAG,QACxD;AAAE;AAAA,MAAQ;AAChB,iBAAW,KAAK,SAAS;AACrB,cAAM,MAAM,OAAO,GAAG,IAAI,IAAI,EAAE,IAAI,KAAK,EAAE;AAC3C,YAAI,EAAE,YAAY,GAAG;AACjB,cAAI,iBAAiB,IAAI,EAAE,IAAI,EAAG;AAClC,eAAU,UAAK,KAAK,EAAE,IAAI,GAAG,GAAG;AAAA,QACpC,WAAW,EAAE,OAAO,GAAG;AACnB,cAAI,KAAK,eAAe,KAAK,MAAM,EAAG;AACtC,gBAAM,MAAW,aAAQ,EAAE,IAAI,EAAE,MAAM,CAAC;AACxC,cAAI,CAAC,cAAc,QAAQ,WAAY,OAAM,KAAK,GAAG;AAAA,QACzD;AAAA,MACJ;AAAA,IACJ,GAfa;AAgBb,SAAK,QAAQ,EAAE;AACf,WAAO;AAAA,EACX;AAAA;AAAA,EAGQ,eAAe,SAAiB,QAA4B;AAChE,QAAI,CAAC,OAAQ,QAAO;AACpB,WAAO,OAAO,KAAK,QAAM;AACrB,YAAM,QAAQ,GACT,QAAQ,SAAS,cAAc,EAC/B,QAAQ,OAAO,UAAU,EACzB,QAAQ,sBAAsB,MAAM,EACpC,QAAQ,qBAAqB,IAAI,EACjC,QAAQ,iBAAiB,OAAO;AACrC,aAAO,IAAI,OAAO,KAAK,EAAE,KAAK,OAAO;AAAA,IACzC,CAAC;AAAA,EACL;AAAA;AAAA,EAGQ,aAAa,YAAoB,SAAiB,MAAuB;AAC7E,UAAM,WAAW,KAAK,IAAI;AAAA,MACtB;AAAA;AAAA;AAAA;AAAA,IAIJ,EAAE,IAAI,YAAY,OAAO;AAEzB,WAAO,SAAS,SAAS,KACrB,SAAS,MAAM,CAAC,MAAW,EAAE,iBAAiB,QAAQ,EAAE,cAAc,IAAI;AAAA,EAClF;AAAA;AAAA,EAGQ,iBAAiB,YAAoB,SAAuB;AAChE,UAAM,YAAY,KAAK,IAAI;AAAA,MACvB;AAAA,IACJ,EAAE,IAAI,YAAY,OAAO;AAEzB,eAAW,OAAO,WAAW;AACzB,WAAK,MAAM,OAAO,IAAI,EAAE;AACxB,WAAK,UAAU,OAAO,IAAI,EAAE;AAAA,IAChC;AACA,SAAK,IAAI;AAAA,MACL;AAAA,IACJ,EAAE,IAAI,YAAY,OAAO;AAAA,EAC7B;AAAA;AAAA,EAGA,MAAc,WACV,YAAoB,SAAiB,SAAiB,MACvC;AACf,UAAM,QAAQ,KAAK,cAAc,SAAS,OAAO;AACjD,UAAM,SAAS,KAAK,YAAY,OAAO;AAEvC,UAAM,cAAc,KAAK,IAAI,QAAQ;AAAA;AAAA;AAAA,SAGpC;AAED,UAAM,WAAqB,CAAC;AAC5B,SAAK,IAAI,YAAY,MAAM;AACvB,eAAS,MAAM,GAAG,MAAM,OAAO,QAAQ,OAAO;AAC1C,cAAM,SAAS,YAAY;AAAA,UACvB;AAAA,UAAY;AAAA,UAAS;AAAA,UAAO,OAAO,GAAG,EAAE;AAAA,UAAM;AAAA,UAAK,OAAO,GAAG,EAAE;AAAA,UAAK;AAAA,QACxE;AACA,iBAAS,KAAK,OAAO,OAAO,eAAe,CAAC;AAAA,MAChD;AAAA,IACJ,CAAC;AAED,UAAM,QAAQ,OAAO,IAAI,OAAK,UAAU,KAAK,YAAY,EAAE,IAAI,EAAE;AACjE,UAAM,aAAa,MAAM,KAAK,WAAW,WAAW,KAAK;AAEzD,UAAM,YAAY,KAAK,IAAI;AAAA,MACvB;AAAA,IACJ;AACA,SAAK,IAAI,YAAY,MAAM;AACvB,eAAS,IAAI,GAAG,IAAI,SAAS,QAAQ,KAAK;AACtC,kBAAU,IAAI,SAAS,CAAC,GAAG,OAAO,KAAK,WAAW,CAAC,EAAE,MAAM,CAAC;AAAA,MAChE;AAAA,IACJ,CAAC;AAED,aAAS,IAAI,GAAG,IAAI,SAAS,QAAQ,KAAK;AACtC,WAAK,MAAM,IAAI,WAAW,CAAC,GAAG,SAAS,CAAC,CAAC;AACzC,WAAK,UAAU,IAAI,SAAS,CAAC,GAAG,WAAW,CAAC,CAAC;AAAA,IACjD;AAEA,WAAO,OAAO;AAAA,EAClB;AAAA;AAAA,EAGA,iBAAiB,YAA0B;AACvC,UAAM,SAAS,KAAK,IAAI;AAAA,MACpB;AAAA,IACJ,EAAE,IAAI,UAAU;AAChB,eAAW,SAAS,QAAQ;AACxB,WAAK,MAAM,OAAO,MAAM,EAAE;AAC1B,WAAK,UAAU,OAAO,MAAM,EAAE;AAAA,IAClC;AAEA,SAAK,IAAI,QAAQ,6CAA6C,EAAE,IAAI,UAAU;AAC9E,SAAK,IAAI,QAAQ,wCAAwC,EAAE,IAAI,UAAU;AACzE,SAAK,IAAI,QAAQ,gDAAgD,EAAE,IAAI,UAAU;AAAA,EACrF;AAAA;AAAA;AAAA,EAKQ,YAAY,MAA+C;AAC/D,QAAI,KAAK,UAAU,cAAc;AAC7B,aAAO,CAAC,EAAE,MAAM,KAAK,EAAE,CAAC;AAAA,IAC5B;AAEA,UAAM,QAAQ,KAAK,MAAM,IAAI;AAC7B,UAAM,cAAc,KAAK,iBAAiB,KAAK;AAC/C,UAAM,SAA0C,CAAC;AACjD,QAAI,aAAa;AAEjB,WAAO,aAAa,KAAK,QAAQ;AAC7B,YAAM,YAAY,KAAK,SAAS;AAChC,UAAI,aAAa,eAAe,cAAc;AAC1C,aAAK,eAAe,MAAM,YAAY,MAAM;AAC5C;AAAA,MACJ;AAEA,YAAM,YAAY,KAAK,eAAe,YAAY,WAAW;AAC7D,YAAM,YAAY,KAAK,MAAM,YAAY,SAAS,EAAE,KAAK;AACzD,UAAI,UAAU,UAAU,iBAAiB;AACrC,eAAO,KAAK,EAAE,MAAM,WAAW,KAAK,WAAW,CAAC;AAAA,MACpD;AACA,mBAAa;AAAA,IACjB;AAEA,WAAO;AAAA,EACX;AAAA;AAAA,EAGQ,eACJ,MAAc,YAAoB,QAC9B;AACJ,UAAM,WAAW,KAAK,MAAM,UAAU,EAAE,KAAK;AAC7C,QAAI,SAAS,UAAU,iBAAiB;AACpC,aAAO,KAAK,EAAE,MAAM,UAAU,KAAK,WAAW,CAAC;AAAA,IACnD,WAAW,OAAO,SAAS,GAAG;AAC1B,aAAO,OAAO,SAAS,CAAC,EAAE,QAAQ,OAAO;AAAA,IAC7C,OAAO;AACH,aAAO,KAAK,EAAE,MAAM,UAAU,KAAK,WAAW,CAAC;AAAA,IACnD;AAAA,EACJ;AAAA;AAAA,EAGQ,eAAe,YAAoB,aAAmC;AAC1E,UAAM,YAAY,aAAa;AAC/B,UAAM,cAAc,YAAY;AAEhC,QAAI,YAAY;AAChB,QAAI,YAAY;AAEhB,eAAW,MAAM,aAAa;AAC1B,UAAI,GAAG,OAAO,WAAY;AAC1B,UAAI,GAAG,MAAM,YAAY,eAAe,EAAG;AAC3C,UAAI,GAAG,MAAM,YAAa;AAE1B,YAAM,WAAW,KAAK,IAAI,GAAG,MAAM,SAAS;AAC5C,YAAM,QAAQ,KAAK,WAAW,iBAAiB,IAAI;AACnD,YAAM,aAAa,GAAG,QAAQ;AAE9B,UAAI,aAAa,WAAW;AACxB,oBAAY;AACZ,oBAAY,GAAG;AAAA,MACnB;AAAA,IACJ;AAEA,WAAO;AAAA,EACX;AAAA;AAAA,EAGQ,iBAAiB,OAA+B;AACpD,UAAM,SAAuB,CAAC;AAC9B,QAAI,UAAU;AACd,QAAI,cAAc;AAElB,eAAW,QAAQ,OAAO;AACtB,UAAI,KAAK,UAAU,EAAE,WAAW,KAAK,GAAG;AACpC,sBAAc,CAAC;AACf,YAAI,CAAC,aAAa;AACd,iBAAO,KAAK,EAAE,KAAK,UAAU,KAAK,SAAS,GAAG,OAAO,GAAG,CAAC;AAAA,QAC7D;AACA,mBAAW,KAAK,SAAS;AACzB;AAAA,MACJ;AAEA,UAAI,aAAa;AACb,mBAAW,KAAK,SAAS;AACzB;AAAA,MACJ;AAEA,iBAAW,CAAC,SAAS,KAAK,KAAK,cAAc;AACzC,YAAI,QAAQ,KAAK,KAAK,KAAK,CAAC,GAAG;AAC3B,iBAAO,KAAK,EAAE,KAAK,SAAS,MAAM,CAAC;AACnC;AAAA,QACJ;AAAA,MACJ;AAEA,iBAAW,KAAK,SAAS;AAAA,IAC7B;AAEA,WAAO;AAAA,EACX;AAAA;AAAA,EAGQ,cAAc,SAAiB,UAA0B;AAC7D,UAAM,QAAQ,QAAQ,MAAM,kBAAkB;AAC9C,QAAI,MAAO,QAAO,MAAM,CAAC,EAAE,KAAK;AAChC,WAAY,cAAS,UAAe,aAAQ,QAAQ,CAAC;AAAA,EACzD;AACJ;;;AC1UA,IAAM,aAAN,MAAoC;AAAA,EAhBpC,OAgBoC;AAAA;AAAA;AAAA,EACvB,OAAO;AAAA,EAChB;AAAA,EACA;AAAA,EACA,WAAW,oBAAI,IAA0B;AAAA,EACjC;AAAA,EACA;AAAA,EAER,MAAM,WAAW,KAAoC;AACjD,SAAK,MAAM,IAAI;AACf,SAAK,aAAa,IAAI;AACtB,SAAK,OAAO,MAAM,IAAI,WAAW;AACjC,QAAI,YAAY,eAAe,YAAY,KAAK,MAAM,KAAK,QAAQ;AACnE,SAAK,UAAU,IAAI,YAAY,IAAI,IAAI,IAAI,WAAW,KAAK,MAAM,KAAK,QAAQ;AAAA,EAClF;AAAA;AAAA,EAGA,cAAc,YAAsC;AAChD,SAAK,IAAI,QAAQ;AAAA;AAAA;AAAA,SAGhB,EAAE;AAAA,MACC,WAAW;AAAA,MACX,WAAW;AAAA,MACX,WAAW,WAAW;AAAA,MACtB,KAAK,UAAU,WAAW,UAAU,CAAC,CAAC;AAAA,MACtC,WAAW,WAAW;AAAA,IAC1B;AAAA,EACJ;AAAA;AAAA,EAGA,iBAAiB,MAAoB;AACjC,SAAK,QAAQ,iBAAiB,IAAI;AAAA,EACtC;AAAA;AAAA,EAGA,kBAAwC;AACpC,WAAQ,KAAK,IAAI,QAAQ,2BAA2B,EAAE,IAAI,EAAY,IAAI,UAAQ;AAAA,MAC9E,MAAM,IAAI;AAAA,MACV,MAAM,IAAI;AAAA,MACV,SAAS,IAAI;AAAA,MACb,QAAQ,KAAK,MAAM,IAAI,WAAW;AAAA,MAClC,SAAS,IAAI;AAAA,IACjB,EAAE;AAAA,EACN;AAAA;AAAA,EAGA,MAAM,iBAAiB,UAGnB,CAAC,GAAkF;AACnF,UAAM,iBAAiB,KAAK,gBAAgB;AAC5C,UAAM,UAAU,QAAQ,cAClB,eAAe,OAAO,OAAK,QAAQ,YAAa,SAAS,EAAE,IAAI,CAAC,IAChE;AAEN,UAAM,UAAgF,CAAC;AAEvF,eAAW,QAAQ,SAAS;AACxB,cAAQ,KAAK,IAAI,IAAI,MAAM,KAAK,QAAQ;AAAA,QACpC,KAAK;AAAA,QACL,KAAK;AAAA,QACL,KAAK;AAAA,QACL;AAAA,UACI,QAAQ,KAAK;AAAA,UACb,YAAY,wBAAC,MAAM,KAAK,UAAU,QAAQ,aAAa,KAAK,MAAM,MAAM,KAAK,KAAK,GAAtE;AAAA,QAChB;AAAA,MACJ;AAAA,IACJ;AAEA,WAAO;AAAA,EACX;AAAA;AAAA,EAGA,MAAM,OAAO,OAAe,SAIA;AACxB,UAAM,IAAI,SAAS,KAAK;AACxB,UAAM,WAAW,MAAM,KAAK,WAAW,MAAM,KAAK;AAIlD,QAAI,UAAU;AACd,QAAI,SAAS,cAAc,KAAK,KAAK,OAAO,GAAG;AAC3C,YAAM,kBAAmB,KAAK,IAAI;AAAA,QAC9B;AAAA,MACJ,EAAE,IAAI,QAAQ,UAAU,GAAW,KAAK;AACxC,YAAM,cAAe,KAAK,IAAI;AAAA,QAC1B;AAAA,MACJ,EAAE,IAAI,GAAW,KAAK;AACtB,YAAM,QAAQ,kBAAkB,IAC1B,KAAK,IAAI,GAAG,KAAK,IAAI,IAAI,KAAK,KAAK,cAAc,eAAe,CAAC,CAAC,IAClE;AACN,gBAAU,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,IAAI;AAAA,IAChD;AAEA,UAAM,OAAO,KAAK,KAAK,OAAO,UAAU,OAAO;AAE/C,UAAM,UAA0B,CAAC;AACjC,eAAW,OAAO,MAAM;AACpB,UAAI,SAAS,YAAY,IAAI,QAAQ,QAAQ,SAAU;AAEvD,YAAM,QAAQ,KAAK,IAAI;AAAA,QACnB;AAAA,MACJ,EAAE,IAAI,IAAI,EAAE;AAEZ,UAAI,CAAC,MAAO;AACZ,UAAI,SAAS,cAAc,MAAM,eAAe,QAAQ,WAAY;AAEpE,YAAM,MAAM,KAAK,eAAe,MAAM,YAAY,MAAM,SAAS;AAEjE,cAAQ,KAAK;AAAA,QACT,MAAM;AAAA,QACN,OAAO,IAAI;AAAA,QACX,UAAU,MAAM;AAAA,QAChB,SAAS,MAAM;AAAA,QACf,SAAS;AAAA,QACT,UAAU;AAAA,UACN,YAAY,MAAM;AAAA,UAClB,OAAO,MAAM;AAAA,UACb,KAAK,MAAM;AAAA,QACf;AAAA,MACJ,CAAC;AAGD,UAAI,QAAQ,UAAU,EAAG;AAAA,IAC7B;AAEA,WAAO;AAAA,EACX;AAAA;AAAA,EAGA,WAAW,YAAoBA,OAAc,SAAuB;AAChE,SAAK,IAAI,QAAQ;AAAA;AAAA;AAAA,SAGhB,EAAE,IAAI,YAAYA,OAAM,OAAO;AAAA,EACpC;AAAA;AAAA,EAGA,cAAc,YAAoBA,OAAoB;AAClD,SAAK,IAAI;AAAA,MACL;AAAA,IACJ,EAAE,IAAI,YAAYA,KAAI;AAAA,EAC1B;AAAA;AAAA,EAGA,eAAwE;AACpE,WAAO,KAAK,IAAI,QAAQ,6BAA6B,EAAE,IAAI;AAAA,EAC/D;AAAA,EAEA,QAA6B;AACzB,WAAO;AAAA,MACH,aAAc,KAAK,IAAI,QAAQ,uCAAuC,EAAE,IAAI,EAAU;AAAA,MACtF,WAAY,KAAK,IAAI,QAAQ,uDAAuD,EAAE,IAAI,EAAU;AAAA,MACpG,QAAS,KAAK,IAAI,QAAQ,sCAAsC,EAAE,IAAI,EAAU;AAAA,MAChF,UAAU,KAAK,KAAK;AAAA,IACxB;AAAA,EACJ;AAAA;AAAA,EAGQ,eAAe,YAAoB,UAAsC;AAC7E,UAAM,QAAQ,SAAS,MAAM,GAAG;AAChC,aAAS,IAAI,MAAM,QAAQ,KAAK,GAAG,KAAK;AACpC,YAAM,YAAY,MAAM,IAAI,MAAM,MAAM,MAAM,MAAM,GAAG,CAAC,EAAE,KAAK,GAAG;AAClE,YAAM,MAAM,KAAK,IAAI;AAAA,QACjB;AAAA,MACJ,EAAE,IAAI,YAAY,SAAS;AAC3B,UAAI,IAAK,QAAO,IAAI;AAAA,IACxB;AAEA,UAAM,OAAO,KAAK,IAAI;AAAA,MAClB;AAAA,IACJ,EAAE,IAAI,UAAU;AAChB,WAAO,MAAM,WAAW;AAAA,EAC5B;AACJ;AAGO,SAAS,OAAgB;AAC5B,SAAO,IAAI,WAAW;AAC1B;AAFgB;","names":["path"]}

package/dist/cli.js

@@ -1,17 +1,17 @@
 #!/usr/bin/env node
 import {
   BrainBank
-} from "./chunk-6MFTQV3O.js";
+} from "./chunk-2BEWWQL2.js";
 import {
   code
-} from "./chunk-WR4WXKJT.js";
+} from "./chunk-MGIFEPYZ.js";
 import {
   git
-} from "./chunk-7JCEW7LT.js";
+} from "./chunk-FI7GWG4W.js";
 import {
   docs
-} from "./chunk-F6SJ3U4H.js";
-import "./chunk-V4UJKXPK.js";
+} from "./chunk-Y3JKI6QN.js";
+import "./chunk-E6WQM4DN.js";
 import "./chunk-QNHBCOKB.js";
 import {
   __name
@@ -37,6 +37,36 @@ function hasFlag(name) {
   return args.includes(`--${name}`);
 }
 __name(hasFlag, "hasFlag");
+var VALUE_FLAGS = /* @__PURE__ */ new Set([
+  "repo",
+  "depth",
+  "collection",
+  "pattern",
+  "context",
+  "name",
+  "keep",
+  "reranker",
+  "only",
+  "docs",
+  "ignore",
+  "meta",
+  "k",
+  "mode",
+  "limit"
+]);
+function stripFlags(argv) {
+  const result = [];
+  for (let i = 0; i < argv.length; i++) {
+    if (argv[i].startsWith("--")) {
+      const name = argv[i].slice(2);
+      if (VALUE_FLAGS.has(name)) i++;
+      continue;
+    }
+    result.push(argv[i]);
+  }
+  return result;
+}
+__name(stripFlags, "stripFlags");
 function printResults(results) {
   if (results.length === 0) {
     console.log(c.yellow("  No results found."));
@@ -145,49 +175,48 @@ async function createBrain(repoPath) {
   const config = await loadConfig();
   const folderIndexers = await discoverFolderIndexers();
   const brainOpts = { repoPath: rp, ...config?.brainbank ?? {} };
+  await setupProviders(brainOpts);
+  const brain = new BrainBank(brainOpts);
+  const builtins = config?.builtins ?? ["code", "git", "docs"];
+  registerBuiltins(brain, rp, builtins);
+  for (const indexer of folderIndexers) brain.use(indexer);
+  if (config?.indexers) {
+    for (const indexer of config.indexers) brain.use(indexer);
+  }
+  return brain;
+}
+__name(createBrain, "createBrain");
+async function setupProviders(brainOpts) {
   const rerankerFlag = getFlag("reranker");
   if (rerankerFlag === "qwen3") {
     const { Qwen3Reranker } = await import("@brainbank/reranker");
     brainOpts.reranker = new Qwen3Reranker();
   }
-  const embeddingEnv = process.env.BRAINBANK_EMBEDDING;
-  if (embeddingEnv === "openai") {
-    const { OpenAIEmbedding } = await import("./openai-CYDMYX7X.js");
+  if (process.env.BRAINBANK_EMBEDDING === "openai") {
+    const { OpenAIEmbedding } = await import("./openai-embedding-VQZCZQYT.js");
     const provider = new OpenAIEmbedding();
     brainOpts.embeddingProvider = provider;
     brainOpts.embeddingDims = provider.dims;
   }
-  const brain = new BrainBank(brainOpts);
-  const builtins = config?.builtins ?? ["code", "git", "docs"];
+}
+__name(setupProviders, "setupProviders");
+function registerBuiltins(brain, rp, builtins) {
   const resolvedRp = path.resolve(rp);
   const hasRootGit = fs.existsSync(path.join(resolvedRp, ".git"));
   const gitSubdirs = !hasRootGit ? detectGitSubdirs(resolvedRp) : [];
   if (gitSubdirs.length > 0 && (builtins.includes("code") || builtins.includes("git"))) {
     console.log(c.cyan(`  Multi-repo: found ${gitSubdirs.length} git repos: ${gitSubdirs.map((d) => d.name).join(", ")}`));
     for (const sub of gitSubdirs) {
-      if (builtins.includes("code")) {
-        brain.use(code({ repoPath: sub.path, name: `code:${sub.name}` }));
-      }
-      if (builtins.includes("git")) {
-        brain.use(git({ repoPath: sub.path, name: `git:${sub.name}` }));
-      }
+      if (builtins.includes("code")) brain.use(code({ repoPath: sub.path, name: `code:${sub.name}` }));
+      if (builtins.includes("git")) brain.use(git({ repoPath: sub.path, name: `git:${sub.name}` }));
     }
   } else {
     if (builtins.includes("code")) brain.use(code({ repoPath: rp }));
     if (builtins.includes("git")) brain.use(git());
   }
   if (builtins.includes("docs")) brain.use(docs());
-  for (const indexer of folderIndexers) {
-    brain.use(indexer);
-  }
-  if (config?.indexers) {
-    for (const indexer of config.indexers) {
-      brain.use(indexer);
-    }
-  }
-  return brain;
 }
-__name(createBrain, "createBrain");
+__name(registerBuiltins, "registerBuiltins");
 
 // src/cli/commands/index-cmd.ts
 async function cmdIndex() {
@@ -454,7 +483,7 @@ async function cmdDocs() {
 }
 __name(cmdDocs, "cmdDocs");
 async function cmdDocSearch() {
-  const query = args.slice(1).filter((a) => !a.startsWith("--")).join(" ");
+  const query = stripFlags(args).slice(1).join(" ");
   if (!query) {
     console.log(c.red("Usage: brainbank dsearch <query>"));
     process.exit(1);
@@ -485,7 +514,7 @@ __name(cmdDocSearch, "cmdDocSearch");
 
 // src/cli/commands/search.ts
 async function cmdSearch() {
-  const query = args.slice(1).join(" ");
+  const query = stripFlags(args).slice(1).join(" ");
   if (!query) {
     console.log(c.red("Usage: brainbank search <query>"));
     process.exit(1);
@@ -500,7 +529,7 @@ async function cmdSearch() {
 }
 __name(cmdSearch, "cmdSearch");
 async function cmdHybridSearch() {
-  const query = args.slice(1).filter((a) => !a.startsWith("--")).join(" ");
+  const query = stripFlags(args).slice(1).join(" ");
   if (!query) {
     console.log(c.red("Usage: brainbank hsearch <query>"));
     process.exit(1);
@@ -516,7 +545,7 @@ async function cmdHybridSearch() {
 }
 __name(cmdHybridSearch, "cmdHybridSearch");
 async function cmdKeywordSearch() {
-  const query = args.slice(1).filter((a) => !a.startsWith("--")).join(" ");
+  const query = stripFlags(args).slice(1).join(" ");
   if (!query) {
     console.log(c.red("Usage: brainbank ksearch <query>"));
     process.exit(1);
@@ -527,7 +556,7 @@ async function cmdKeywordSearch() {
 \u2501\u2501\u2501 BrainBank Keyword Search: "${query}" \u2501\u2501\u2501`));
   console.log(c.dim(`  Mode: BM25 full-text (instant)
 `));
-  const results = brain.searchBM25(query);
+  const results = await brain.searchBM25(query);
   printResults(results);
   brain.close();
 }
@@ -566,7 +595,7 @@ async function cmdContext() {
     brain2.close();
     return;
   }
-  const task = args.slice(1).join(" ");
+  const task = stripFlags(args).slice(1).join(" ");
   if (!task) {
     console.log(c.red("Usage: brainbank context <task description>"));
     console.log(c.dim("       brainbank context add <collection> <path> <description>"));