Letta
Self-hosted AI agent platform with persistent memory. Local agents connect to Ollama or LM Studio; cloud model connections use remote providers.

Letta is an AI agent platform whose App Server can run locally or on a self-hosted server. It's for people building assistants that retain context between conversations and use that memory in later interactions. Local agents can use Ollama or LM Studio with agent state and the model endpoint on your machine. Cloud model connections use remote providers.
Formerly called MemGPT, Letta focuses on agents with persistent memory whose behavior can develop through experience. Its Letta Agent product applies that approach to an assistant whose memory, identity and capabilities evolve over time.
You can interact with agents through a terminal interface or desktop apps for macOS, Windows and Linux. Slack, Telegram, Discord and custom channels provide other ways to reach them. Letta also offers a separate hosted browser service, accessible through mobile browsers.
The current implementation lives in letta-ai/letta-code and includes the agent runtime, terminal interface, App Server and conversation channels used by the desktop and web apps. The linked letta-ai/letta repository carries the Apache 2.0 license. Its retired V1 API server remains available as historical source, while active development continues in letta-ai/letta-code.