Letta (MemGPT)

framework active open-source

An agent framework focused on long-term memory and stateful AI assistants. Originally known as MemGPT, Letta implements a memory hierarchy that allows agents to manage their own context, maintaining conversations and knowledge across sessions without losing information to context limits.

Implements

Concepts this tool claims to implement:

  • Agent Memory primary

    Hierarchical memory system with main context, archival memory, and recall memory. Agents self-manage memory through function calls. Persistent memory across sessions.

  • Long-Term Memory primary

    Archival memory for unbounded storage. Vector-based retrieval for relevant memories. Agents can write to and search archival memory.

  • Agent State primary

    Persistent agent state including memory, persona, and conversation history. Server-based deployment for stateful agents.

  • Agent Orchestration secondary

    Agent server for deploying and managing agents. Tool support for extending agent capabilities. Multi-agent coordination.

Integration Surfaces

  • Python SDK
  • REST API
  • Agent Development Environment (ADE)
  • Docker deployment

Details

Vendor
Letta Inc.
License
Apache-2.0
Runs On
local, cloud
Used By
human, system

Links

Notes

Letta (formerly MemGPT) pioneered the idea of LLMs managing their own memory through function calls. Key insight: treat context window as working memory, let the agent manage what's in it. Good for building persistent AI assistants that remember across conversations.

Related Tools