promptgraph-mcp 1.2.0 → 1.3.0

package/README.md

@@ -129,7 +129,7 @@ promptgraph-mcp reindex   # Re-index all skills
 | Skills in context | All 40+ | 1 (router) |
 | Tokens per session | ~20,000–50,000 | ~150 + 1 skill |
 
-> **Note:** Search is currently linear O(N) over all chunks. Works well up to ~1,000 skills. ANN index (HNSW/LanceDB) planned for larger collections.
+> **Search:** Uses HNSW index (via [vectra](https://github.com/Stevenic/vectra)) for O(log N) approximate nearest neighbor search. Falls back to brute-force on first run before index is built.
 
 ## File Structure
 

package/ann.js

@@ -0,0 +1,50 @@
+import { LocalIndex } from 'vectra';
+import path from 'path';
+import os from 'os';
+import { getDb } from './db.js';
+
+const INDEX_PATH = path.join(os.homedir(), '.claude', '.promptgraph', 'hnsw-index');
+
+let _index = null;
+
+async function getIndex() {
+  if (_index) return _index;
+  _index = new LocalIndex(INDEX_PATH);
+  if (!await _index.isIndexCreated()) {
+    await _index.createIndex({ version: 1, deleteIfExists: true });
+  }
+  return _index;
+}
+
+export async function buildAnnIndex() {
+  const index = await getIndex();
+  await index.createIndex({ version: 1, deleteIfExists: true });
+
+  const db = getDb();
+  const chunks = db.prepare('SELECT skill_id, chunk_index, embedding FROM chunks').all();
+
+  for (const chunk of chunks) {
+    const vec = JSON.parse(chunk.embedding);
+    await index.insertItem({
+      vector: vec,
+      metadata: { skill_id: chunk.skill_id, chunk_index: chunk.chunk_index },
+    });
+  }
+
+  console.error(`[PromptGraph] ANN index built: ${chunks.length} chunks`);
+}
+
+export async function annSearch(queryVec, topK = 20) {
+  try {
+    const index = await getIndex();
+    if (!await index.isIndexCreated()) return null;
+
+    const results = await index.queryItems(queryVec, topK);
+    return results.map(r => ({
+      skill_id: r.item.metadata.skill_id,
+      score: r.score,
+    }));
+  } catch {
+    return null;
+  }
+}

package/indexer.js

@@ -4,6 +4,7 @@ import { embed } from './embedder.js';
 import { getDb, skillId } from './db.js';
 import { loadConfig } from './config.js';
 import { chunkText } from './chunker.js';
+import { buildAnnIndex } from './ann.js';
 
 async function indexSkill(db, skill) {
   const id = skillId(skill.source, skill.name);
@@ -53,6 +54,8 @@ export async function indexAll() {
       }
     }
   }
+  process.stdout.write('\r  Building ANN index...');
+  await buildAnnIndex();
   console.log(`\n  Done. ${count} skills indexed.`);
 }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "promptgraph-mcp",
-  "version": "1.2.0",
+  "version": "1.3.0",
   "main": "index.js",
   "type": "module",
   "bin": {
@@ -37,6 +37,7 @@
     "chokidar": "^5.0.0",
     "fastembed": "^2.1.0",
     "glob": "^13.0.6",
-    "gray-matter": "^4.0.3"
+    "gray-matter": "^4.0.3",
+    "vectra": "^0.15.0"
   }
 }

package/search.js

@@ -1,34 +1,52 @@
 import { embed, cosineSimilarity } from './embedder.js';
 import { getDb } from './db.js';
+import { annSearch } from './ann.js';
 
 export async function search(query, topK = 5) {
   const db = getDb();
   const queryVec = await embed(query);
 
-  const chunks = db.prepare('SELECT skill_id, embedding FROM chunks').all();
+  // Try ANN index first (fast, O(log N))
+  const annResults = await annSearch(queryVec, topK * 4);
+
+  if (annResults && annResults.length > 0) {
+    const bestBySkill = new Map();
+    for (const r of annResults) {
+      const prev = bestBySkill.get(r.skill_id);
+      if (!prev || r.score > prev) bestBySkill.set(r.skill_id, r.score);
+    }
+    return [...bestBySkill.entries()]
+      .sort((a, b) => b[1] - a[1])
+      .slice(0, topK)
+      .map(([id, score]) => {
+        const skill = db.prepare('SELECT id, name, description, path, source FROM skills WHERE id = ?').get(id);
+        return skill ? { ...skill, score } : null;
+      })
+      .filter(Boolean);
+  }
 
+  // Fallback: brute force (used before first reindex)
+  const chunks = db.prepare('SELECT skill_id, embedding FROM chunks').all();
   const bestBySkill = new Map();
   for (const chunk of chunks) {
     const score = cosineSimilarity(queryVec, JSON.parse(chunk.embedding));
     const prev = bestBySkill.get(chunk.skill_id);
     if (!prev || score > prev) bestBySkill.set(chunk.skill_id, score);
   }
-
-  const topIds = [...bestBySkill.entries()]
+  return [...bestBySkill.entries()]
     .sort((a, b) => b[1] - a[1])
     .slice(0, topK)
-    .map(([id]) => id);
-
-  return topIds.map(id => {
-    const skill = db.prepare('SELECT id, name, description, path, source FROM skills WHERE id = ?').get(id);
-    return { ...skill, score: bestBySkill.get(id) };
-  });
+    .map(([id, score]) => {
+      const skill = db.prepare('SELECT id, name, description, path, source FROM skills WHERE id = ?').get(id);
+      return skill ? { ...skill, score } : null;
+    })
+    .filter(Boolean);
 }
 
 export function getContext(id) {
   const db = getDb();
   const skill = db.prepare('SELECT * FROM skills WHERE id = ?').get(id)
-    || db.prepare("SELECT * FROM skills WHERE name = ? ORDER BY id LIMIT 1").get(id);
+    || db.prepare('SELECT * FROM skills WHERE name = ? ORDER BY id LIMIT 1').get(id);
   if (!skill) return null;
   const callees = db.prepare('SELECT to_skill FROM edges WHERE from_skill = ?').all(skill.id).map(r => r.to_skill);
   const callers = db.prepare('SELECT from_skill FROM edges WHERE to_skill = ?').all(skill.id).map(r => r.from_skill);