@engram-ai-memory/core

The Brain — memory engine, embeddings, knowledge graph, and retrieval for Engram.

Install

npm install @engram-ai-memory/core

Quick Start

import { NeuralBrain } from '@engram-ai-memory/core';

const brain = new NeuralBrain({
  dbPath: './engram.db',
  defaultSource: 'my-app',
});

await brain.initialize();

// Store a memory
const { memory } = await brain.store({
  content: 'User prefers TypeScript',
  type: 'semantic',
  importance: 0.8,
});

// Recall context
const result = await brain.recall('What language does the user prefer?');
console.log(result.context);

// Search
const results = await brain.search('TypeScript');

// Stats
const stats = await brain.stats();

brain.shutdown();

Features

• 3 Memory Types — Episodic (events), Semantic (facts + knowledge graph), Procedural (patterns)
• 7-Step Recall — Embed, vector search, graph expand, score, rank, truncate, log
• Memory Decay — Ebbinghaus forgetting curve with auto-consolidation
• Namespace Isolation — Scoped memory workspaces per project/agent
• Contradiction Detection — Auto-detect conflicting memories with resolution strategies
• Embedding Upgradability — Swap models, re-embed pipeline, model tracking
• Streaming Recall — Progressive context assembly via async generator
• Index Persistence — Save/load vector index for fast startup (27x speedup)
• Webhooks — Subscribe to memory events (stored, forgotten, decayed, consolidated, contradiction)
• Tags & Collections — Tag cloud, filter by tag, prefix-based collections
• Plugin System — 6 lifecycle hooks (onStore, onRecall, onForget, onDecay, onStartup, onShutdown)
• PostgreSQL Support — Optional, via ENGRAM_DATABASE=postgresql

Environment Variables

| Variable | Default | Description | |---|---|---| | ENGRAM_DB_PATH | ./engram.db | SQLite database path | | ENGRAM_DATABASE | sqlite | sqlite or postgresql | | DATABASE_URL | — | PostgreSQL connection URL | | ENGRAM_NAMESPACE | — | Memory namespace isolation | | ENGRAM_INDEX_PATH | {dbPath}.index | Persistent vector index path | | ENGRAM_EMBEDDING_MODEL | Xenova/all-MiniLM-L6-v2 | Embedding model |

Embedding

Uses Xenova/all-MiniLM-L6-v2 (384-dim) via ONNX/WASM — runs locally, no API calls, no cost. FP16 compressed for 2x storage savings.

Links

• GitHub
• Documentation
• Website

License

MIT