promptgraph-mcp 2.1.3 → 2.1.4

package/chunker.js

@@ -1,6 +1,6 @@
 const CHUNK_SIZE = 800;
 const CHUNK_OVERLAP = 100;
-const MAX_CHUNKS = 3;
+const MAX_CHUNKS = 2;
 
 export function chunkText(text) {
   // Split on markdown h1/h2/h3 headers to preserve semantic boundaries

package/db.js

@@ -46,6 +46,15 @@ export function getDb() {
       success INTEGER DEFAULT 0,
       fail INTEGER DEFAULT 0
     );
+
+    CREATE VIRTUAL TABLE IF NOT EXISTS skills_fts USING fts5(
+      id UNINDEXED,
+      name,
+      description,
+      content,
+      content='skills',
+      content_rowid='rowid'
+    );
   `);
 
   // migrate: add hash column if missing

package/embedder.js

@@ -3,7 +3,7 @@ import path from 'path';
 import os from 'os';
 
 const CACHE_DIR = path.join(os.homedir(), '.claude', '.promptgraph', 'model-cache');
-const BATCH_SIZE = 64;
+const BATCH_SIZE = 256;
 
 let model = null;
 

package/index.js

@@ -565,7 +565,9 @@ if (args[0] === 'update') {
 
 if (args[0] === 'reindex') {
   const { indexAll } = await import('./indexer.js');
-  await indexAll();
+  const fast = args.includes('--fast');
+  if (fast) info(chalk.yellow('Fast mode — skipping embeddings (keyword search only)'));
+  await indexAll({ fast });
   process.exit(0);
 }
 

package/indexer.js

@@ -11,7 +11,7 @@ import { buildAnnIndex } from './ann.js';
 import { progress, progressDone, success, info, spinner } from './cli.js';
 import chalk from 'chalk';
 
-async function indexBatch(db, skills) {
+async function indexBatch(db, skills, { fast = false } = {}) {
   const upsertSkill = db.prepare(`
     INSERT INTO skills (id, name, description, path, source, content, hash)
     VALUES (@id, @name, @description, @path, @source, @content, @hash)
@@ -26,32 +26,43 @@ async function indexBatch(db, skills) {
   const deleteEdges = db.prepare('DELETE FROM edges WHERE from_skill = ?');
   const upsertChunk = db.prepare('INSERT OR REPLACE INTO chunks (skill_id, chunk_index, text, embedding) VALUES (?, ?, ?, ?)');
   const upsertEdge = db.prepare('INSERT OR IGNORE INTO edges (from_skill, to_skill) VALUES (?, ?)');
+  const upsertFts = db.prepare(`INSERT OR REPLACE INTO skills_fts(id, name, description, content) VALUES (?, ?, ?, ?)`);
 
   const allChunks = [];
-  for (const skill of skills) {
-    const id = skillId(skill.source, skill.name);
-    const chunks = chunkText(skill.name + ' ' + skill.description + '\n' + skill.content);
-    for (let i = 0; i < chunks.length; i++) {
-      allChunks.push({ id, skill, chunkIndex: i, text: chunks[i] });
+  if (!fast) {
+    for (const skill of skills) {
+      const id = skillId(skill.source, skill.name);
+      const chunks = chunkText(skill.name + ' ' + skill.description + '\n' + skill.content);
+      for (let i = 0; i < chunks.length; i++) {
+        allChunks.push({ id, skill, chunkIndex: i, text: chunks[i] });
+      }
     }
   }
 
-  const texts = allChunks.map(c => c.text);
-  process.stdout.write(`  Embedding ${texts.length} chunks...`);
-  const embeddings = await embedBatch(texts);
-  process.stdout.write('\r' + ' '.repeat(40) + '\r');
+  let embeddings = [];
+  if (!fast && allChunks.length) {
+    const texts = allChunks.map(c => c.text);
+    process.stdout.write(`  Embedding ${texts.length} chunks...`);
+    embeddings = await embedBatch(texts);
+    process.stdout.write('\r' + ' '.repeat(40) + '\r');
+  }
 
   // pass 1: upsert all skills + chunks (no edges yet)
   db.transaction(() => {
     for (const skill of skills) {
       const id = skillId(skill.source, skill.name);
       upsertSkill.run({ id, name: skill.name, description: skill.description, path: skill.path, source: skill.source, content: skill.content, hash: skill.hash || null });
-      deleteChunks.run(id);
-      deleteEdges.run(id);
+      upsertFts.run(id, skill.name, skill.description || '', skill.content || '');
+      if (!fast) {
+        deleteChunks.run(id);
+        deleteEdges.run(id);
+      }
     }
-    for (let i = 0; i < allChunks.length; i++) {
-      const { id, chunkIndex, text } = allChunks[i];
-      upsertChunk.run(id, chunkIndex, text, vecToBlob(embeddings[i]));
+    if (!fast) {
+      for (let i = 0; i < allChunks.length; i++) {
+        const { id, chunkIndex, text } = allChunks[i];
+        upsertChunk.run(id, chunkIndex, text, vecToBlob(embeddings[i]));
+      }
     }
   })();
 
@@ -71,7 +82,7 @@ async function indexBatch(db, skills) {
   })();
 }
 
-export async function indexAll() {
+export async function indexAll({ fast = false } = {}) {
   const config = loadConfig();
   const db = getDb();
 
@@ -163,7 +174,7 @@ export async function indexAll() {
       batch.push({ ...parsed, hash });
 
       if (batch.length >= BATCH_SIZE) {
-        await indexBatch(db, batch);
+        await indexBatch(db, batch, { fast });
         count += batch.length;
         batch = [];
         const eta = count > 0 ? Math.round((total - count) * (Date.now() - start) / count / 1000) : '?';
@@ -185,17 +196,20 @@ export async function indexAll() {
   }
 
   if (batch.length > 0) {
-    await indexBatch(db, batch);
+    await indexBatch(db, batch, { fast });
     count += batch.length;
   }
 
   progress(total, total, { skipped, errors });
   progressDone();
-  const spin = spinner('Building ANN index...');
-  spin.start();
-  await buildAnnIndex();
-  spin.stop();
+  if (!fast) {
+    const spin = spinner('Building ANN index...');
+    spin.start();
+    await buildAnnIndex();
+    spin.stop();
+  }
   success(`Indexed ${chalk.white.bold(count)} skills  ${chalk.gray(`(${errors} errors, ${skipped} skipped, ${removed} removed)`)}`);
+  if (fast) info(chalk.yellow('Fast mode: keyword search only. Run `pg reindex` for semantic search.'));
   const elapsed = ((Date.now() - start) / 1000).toFixed(1);
   info(chalk.gray(`Time: ${elapsed}s`));
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "promptgraph-mcp",
-  "version": "2.1.3",
+  "version": "2.1.4",
   "main": "index.js",
   "type": "module",
   "bin": {

package/search.js

@@ -32,20 +32,35 @@ export async function search(query, topK = 5) {
       .filter(Boolean);
   }
 
-  // Fallback: brute force (used before first reindex)
+  // Fallback: brute force cosine (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, blobToVec(chunk.embedding));
-    const prev = bestBySkill.get(chunk.skill_id);
-    if (!prev || score > prev) bestBySkill.set(chunk.skill_id, score);
+  if (chunks.length > 0) {
+    const bestBySkill = new Map();
+    for (const chunk of chunks) {
+      const score = cosineSimilarity(queryVec, blobToVec(chunk.embedding));
+      const prev = bestBySkill.get(chunk.skill_id);
+      if (!prev || score > prev) bestBySkill.set(chunk.skill_id, score);
+    }
+    return [...bestBySkill.entries()]
+      .map(([id, score]) => ({ id, score: applyRatingBoost(db, id, score) }))
+      .sort((a, b) => b.score - a.score)
+      .slice(0, topK)
+      .map(({ id, score }) => skillWithSnippet(db, id, score))
+      .filter(Boolean);
+  }
+
+  // Fast-mode fallback: FTS5 keyword search (no embeddings)
+  try {
+    const terms = query.replace(/[^\w\s]/g, ' ').trim().split(/\s+/).filter(Boolean).join(' OR ');
+    const rows = db.prepare(
+      `SELECT s.id, bm25(skills_fts) AS score FROM skills_fts
+       JOIN skills s ON skills_fts.id = s.id
+       WHERE skills_fts MATCH ? ORDER BY score LIMIT ?`
+    ).all(terms, topK);
+    return rows.map(r => skillWithSnippet(db, r.id, Math.max(0, 1 + r.score / 10))).filter(Boolean);
+  } catch {
+    return [];
   }
-  return [...bestBySkill.entries()]
-    .map(([id, score]) => ({ id, score: applyRatingBoost(db, id, score) }))
-    .sort((a, b) => b.score - a.score)
-    .slice(0, topK)
-    .map(({ id, score }) => skillWithSnippet(db, id, score))
-    .filter(Boolean);
 }
 
 function skillWithSnippet(db, id, score) {