claude-mem

A lightweight memory management utility for Claude and other LLM-based agents. Claude-mem provides persistent, structured memory that survives across conversation sessions, enabling agents to recall context, preferences, and learned information.

Features

• Persistent Memory: Store and retrieve conversation context across sessions
• Structured Storage: Organize memories by topic, project, or custom taxonomy
• Semantic Search: Find relevant memories using embedding-based similarity search
• Memory Decay: Configurable time-based relevance scoring so stale memories fade
• Multi-Agent Support: Shared memory pools for collaborative agent workflows
• Privacy Controls: Local-first storage with optional encrypted cloud sync

Installation

npm install claude-mem

Quick Start

import { MemoryStore } from 'claude-mem';

const memory = new MemoryStore({
  storagePath: '.claude-mem',
  maxMemories: 10000,
  decayFactor: 0.95,
});

// Store a memory
await memory.store({
  content: 'User prefers TypeScript over JavaScript',
  tags: ['preference', 'language'],
  importance: 0.8,
});

// Retrieve relevant memories
const results = await memory.search('What language does the user prefer?', {
  limit: 5,
  minRelevance: 0.3,
});

Configuration

Architecture

Claude-mem uses a hybrid storage approach:

• Short-term: In-memory LRU cache for active session context
• Long-term: SQLite database with vector extensions for persistent storage
• Index: HNSW index for fast approximate nearest neighbor search

API Reference

See the full API documentation for detailed usage.

License

MIT License