Letta — AI Agent Long-Term Memory Framework

What is Letta?

Letta (formerly MemGPT) is a framework for building AI agents with long-term memory. It breaks through context window limits using a tiered memory architecture (core / recall / archival), with the agent autonomously managing its memory.

In one sentence: AI agent long-term memory framework — tiered memory architecture breaks through context limits, the agent decides what to remember and forget — 12k+ stars.

For: Developers building AI agents that need cross-session memory.

Core Features

1. Tiered Memory

Core memory (always in context), recall memory (conversation history), archival memory (unlimited storage).

2. Agent-Managed Memory

The agent autonomously decides which information to store in long-term memory.

3. Tool Calling

Supports custom tools defined via Python decorators.

4. REST API

Built-in server with a complete REST API for managing agents.

FAQ

Q: How is it different from RAG? A: RAG retrieves from static documents; Letta agents actively manage their own memory.

Q: Does it support local models? A: Yes — Ollama, vLLM, and others.