Summary
Letta (formerly MemGPT) is a platform for building stateful agents. It exposes a layered memory system, memory blocks and tool integrations that let agents learn and self-improve, suitable for research and production.
Key features
- Hierarchical memory model (memory blocks) with support for persistence and edits.
- SDKs for Python and TypeScript, an Agent Development Environment (ADE) for no-code agent creation, and a Letta Desktop for local use.
- Support for local model providers (Ollama, LM Studio), MCP integration and custom tools.
Use cases
- Long-running, self-improving agents for customer support, research assistants and workflow automation.
- Multi-agent systems with shared memory and persistent agent state management.
Technical details
- Python-first core service with a TypeScript client, plugin-style tools and support for Agent File (.af) export/import and background/async execution modes.