Provider pricing
Letta AI API pricing
Letta (formerly MemGPT, from the UC Berkeley research project) is a stateful agent runtime that gives any LLM persistent long-term memory, tool use, and the ability to self-edit its own context. The hosted Letta Cloud is priced per agent-step with free dev usage; the open-source server is Apache 2.0 and works with OpenAI, Anthropic, Google, and any local model.
Founded2023
HeadquartersSan Francisco, CA
Tracked modelsDirectory
Best for
Builders of long-running personal assistants, customer-support agents, or research copilots that must remember across sessions.
Strengths
- Persistent agent memory — core, archival, and recall layers managed by the LLM itself
- Provider-agnostic — swap OpenAI, Claude, Gemini, or local models per agent
- ADE (Agent Development Environment) for inspecting and editing agent state visually
- Apache 2.0 server keeps self-hosted deployments fully unlocked