@4meta5/semantic-memory

Standalone semantic memory storage with hybrid search.

Installation

npm install @4meta5/semantic-memory

Usage

import { createMemoryStore } from '@4meta5/semantic-memory';

const store = createMemoryStore({ dbPath: './memory.db' });

// Add memories
await store.add('JWT tokens for auth', { topics: ['auth'] });
await store.add('Use Vitest for testing', { topics: ['testing'] });

// Search with BM25
const results = store.searchBM25('authentication', 10);

// Get stats
console.log(store.getStats());

store.close();

Features

• BM25 full-text search via SQLite FTS5
• Optional vector search with pluggable embedding providers
• RRF fusion for hybrid ranking
• Local-first SQLite storage

API

createMemoryStore(config)

Create a memory store instance.

interface MemoryStoreConfig {
  dbPath: string;                    // Path to SQLite database
  embeddingProvider?: EmbeddingProvider;  // Optional for vector search
  useVectors?: boolean;              // Enable vector search
}

store.add(content, metadata?, options?)

Add a memory to the store.

store.searchBM25(query, limit)

Search using BM25 full-text ranking.

store.searchHybrid(query, limit)

Search using hybrid BM25 + vector with RRF fusion. Requires embedding provider.

store.getStats()

Get storage statistics.

License

MIT