Letta

Letta grew out of the MemGPT research project and has become a platform for building stateful agents: agents with persistent, editable memory that carries across sessions and lets them learn over time instead of starting fresh with every conversation. The framework is model agnostic, working with Anthropic, OpenAI, and other providers, and supports skills, subagents, and tool use. Developers can work through the TypeScript Agent SDK, older Python and TypeScript SDKs that target the Letta API directly, or Letta Code, a CLI for running agents in the terminal. Deployment options span a self hosted server, a desktop application, a hosted cloud platform, and channel integrations such as Slack. The core server is open source and maintained by a community of more than a hundred contributors. Teams building assistants that need durable long term memory, including coding agents, personal assistants, and customer facing bots that remember their users, are the typical adopters.