Letta (formerly MemGPT) builds stateful agents with long-term memory that learn from experience and improve with use. Its open-source agent harness draws on the Letta lab's research into AI memory and continual learning, and it is used for coding agents, personal assistants, AI coworkers, and stateful agents embedded in applications.
Active development now lives in Letta Code, which contains the agent harness, an interactive terminal UI, the App Server for local or self-hosted agents, chat channels, and the runtime behind the desktop and web apps. The original Letta V1 API server is retired and kept on an archive branch for reproducibility.
Agents are reachable from a terminal UI, a desktop app for macOS, Windows, and Linux, a web chat that also works on mobile, and Slack, Telegram, Discord, or custom channels. Letta Cloud keeps an agent's memory, identity, and conversations available across computers, and the Letta Agent SDK embeds agents in TypeScript applications.
Features
- Persistent memory: agents keep memory, identity, and conversations across sessions and can customize their own memory system
- Skills: learned or pre-made skills extend what an agent can do
- Schedules: recurring prompts automate repeated work
- Channels: agents answer in Slack, Telegram, Discord, or custom channels
- Agent SDK: a TypeScript SDK for building stateful agents into applications
- Deployment choice: managed hosting on Letta Cloud or the full runtime self-hosted through the App Server
- Mods: extensions that change the harness itself
- GitHub Action: runs a Letta coding agent from a GitHub workflow
