Letta

Letta is a platform for building personalized AI agents that maintain continuous identity and memory across interactions. Agents can learn from experience by rewriting their own context, skills, and prompts without requiring model updates. The system supports deployment across multiple channels including Slack, Discord, Telegram, and custom integrations, as well as local and cloud environments.

Designed for teams seeking AI assistants that improve through use, Letta offers features like scheduled task execution, multi-device support, and git-based memory versioning through MemFS. Agents can work proactively, search conversation history, and configure themselves. The project originated from MemGPT research at UC Berkeley's Sky Computing Lab and is built using Claude and Cursor tools.