@mclean-capital/neura 2.4.2 → 3.1.0

package/stores/index.js

@@ -4385,8 +4385,14 @@ var Logger = class _Logger {
 
 // src/stores/migrations.ts
 var log = new Logger("store");
-async function runMigrations(db) {
+async function runMigrations(db, embeddingDimensions = 3072) {
   await db.exec("CREATE EXTENSION IF NOT EXISTS vector;");
+  await db.exec(`
+    CREATE TABLE IF NOT EXISTS _meta (
+      key TEXT PRIMARY KEY,
+      value TEXT NOT NULL
+    )
+  `);
   await db.exec(`
     CREATE TABLE IF NOT EXISTS sessions (
       id TEXT PRIMARY KEY,
@@ -4437,7 +4443,7 @@ async function runMigrations(db) {
       content TEXT NOT NULL,
       category TEXT NOT NULL DEFAULT 'general',
       tags JSONB NOT NULL DEFAULT '[]',
-      embedding vector(3072),
+      embedding vector(${embeddingDimensions}),
       source_session_id TEXT,
       confidence REAL NOT NULL DEFAULT 0.8,
       access_count INTEGER NOT NULL DEFAULT 0,
@@ -4452,10 +4458,13 @@ async function runMigrations(db) {
     `SELECT format_type(atttypid, atttypmod) AS col_type FROM pg_attribute
      WHERE attrelid = 'facts'::regclass AND attname = 'embedding'`
   );
-  if (typeCheck.rows.length > 0 && typeCheck.rows[0].col_type !== "vector(3072)") {
+  if (typeCheck.rows.length > 0 && typeCheck.rows[0].col_type !== `vector(${embeddingDimensions})`) {
     await db.exec("ALTER TABLE facts DROP COLUMN embedding");
-    await db.exec("ALTER TABLE facts ADD COLUMN embedding vector(3072)");
-    log.info("migrated facts.embedding to vector(3072)", { was: typeCheck.rows[0].col_type });
+    await db.exec(`ALTER TABLE facts ADD COLUMN embedding vector(${embeddingDimensions})`);
+    log.info("migrated facts.embedding", {
+      was: typeCheck.rows[0].col_type,
+      now: `vector(${embeddingDimensions})`
+    });
   }
   await db.exec(
     "CREATE UNIQUE INDEX IF NOT EXISTS idx_facts_content_category ON facts(content, category)"
@@ -4540,7 +4549,9 @@ async function runMigrations(db) {
       content || ' ' || COALESCE((SELECT string_agg(value, ' ') FROM jsonb_array_elements_text(tags)), '')
     ) WHERE tsv IS NULL
   `);
-  await db.exec("ALTER TABLE transcripts ADD COLUMN IF NOT EXISTS embedding vector(3072)");
+  await db.exec(
+    `ALTER TABLE transcripts ADD COLUMN IF NOT EXISTS embedding vector(${embeddingDimensions})`
+  );
   await db.exec(`
     CREATE TABLE IF NOT EXISTS entities (
       id TEXT PRIMARY KEY,
@@ -4583,7 +4594,7 @@ async function runMigrations(db) {
       id TEXT PRIMARY KEY,
       session_id TEXT NOT NULL REFERENCES sessions(id),
       chunk_text TEXT NOT NULL,
-      embedding vector(3072),
+      embedding vector(${embeddingDimensions}),
       start_transcript_id INTEGER NOT NULL,
       end_transcript_id INTEGER NOT NULL,
       created_at TIMESTAMP NOT NULL DEFAULT NOW()
@@ -4621,6 +4632,29 @@ async function runMigrations(db) {
       use_count INTEGER NOT NULL DEFAULT 0
     )
   `);
+  const storedDims = await db.query(
+    `SELECT value FROM _meta WHERE key = 'embedding_dimensions'`
+  );
+  if (storedDims.rows.length === 0) {
+    await db.query(`INSERT INTO _meta (key, value) VALUES ('embedding_dimensions', $1)`, [
+      String(embeddingDimensions)
+    ]);
+  } else if (storedDims.rows[0].value !== String(embeddingDimensions)) {
+    log.warn("embedding dimensions changed", {
+      was: storedDims.rows[0].value,
+      now: embeddingDimensions
+    });
+    await db.query(`UPDATE _meta SET value = $1 WHERE key = 'embedding_dimensions'`, [
+      String(embeddingDimensions)
+    ]);
+    for (const table of ["facts", "transcripts", "transcript_chunks"]) {
+      await db.exec(`ALTER TABLE ${table} DROP COLUMN IF EXISTS embedding`);
+      await db.exec(`ALTER TABLE ${table} ADD COLUMN embedding vector(${embeddingDimensions})`);
+    }
+    log.info("embedding columns recreated with new dimensions", {
+      dimensions: embeddingDimensions
+    });
+  }
   const identityCount = await db.query(
     "SELECT COUNT(*)::TEXT as count FROM identity"
   );
@@ -4833,7 +4867,7 @@ async function getFacts(db, options = {}) {
   return result.rows.map((r) => mapFact(r));
 }
 async function searchFacts(db, query, embedding, limit = 10) {
-  if (embedding?.length === 3072) {
+  if (embedding && embedding.length > 0) {
     const vecStr = `[${embedding.join(",")}]`;
     const result2 = await db.query(
       `SELECT * FROM facts
@@ -4858,9 +4892,6 @@ async function searchFacts(db, query, embedding, limit = 10) {
   return result.rows.map((r) => mapFact(r));
 }
 async function upsertFact(db, content, category, tags, sourceSessionId, confidence = 0.8, embedding, tagPath) {
-  if (embedding && embedding.length !== 3072) {
-    throw new Error(`Embedding must be 3072-dimensional, got ${embedding.length}`);
-  }
   const id = crypto3.randomUUID();
   const embeddingStr = embedding ? `[${embedding.join(",")}]` : null;
   const resolvedTagPath = tagPath ?? category;
@@ -5754,12 +5785,12 @@ var PgliteStore = class _PgliteStore {
    * Create a PgliteStore backed by a directory on disk (WAL-persisted).
    * If no dataDir is provided, creates an in-memory instance (useful for tests).
    */
-  static async create(dataDir) {
+  static async create(dataDir, embeddingDimensions) {
     const db = await PGlite.create(dataDir ?? "memory://", {
       extensions: { vector }
     });
     const store = new _PgliteStore(db);
-    await runMigrations(db);
+    await runMigrations(db, embeddingDimensions);
     return store;
   }
   // --- Session methods ---