Letta Code

Letta (formerly MemGPT) provides a stateful control plane for LLM agents with persistent memory and tool-use capabilities.

Letta manages the complex orchestration of Long-term Memory for AI agents. By implementing a tiered storage architecture (similar to an OS), it allows agents to store, retrieve, and update information across sessions without hitting context window limits. Developers use the Letta framework to build autonomous systems that maintain consistent personas and evolve based on user interactions. The core engine handles the recursive loops required for self-editing memory and precise tool execution: ensuring agents remain grounded in historical data while operating in real-time environments.

https://github.com/letta-ai/letta