Letta
Framework for building stateful AI agents with persistent memory, formerly MemGPT
Letta, formerly MemGPT, is a platform for building AI agents with persistent memory. Originating from UC Berkeley research, it implements an OS-inspired memory architecture where agents manage core memory (in-context, like RAM) and archival memory (external storage, like disk). This lets agents learn and self-improve across conversations rather than being stateless. Agent state is persisted in databases rather than Python variables. Available as open-source (Apache 2.0) for self-hosting, or as Letta Cloud with a free tier of 5,000 monthly credits.
Pricing: Free / monthly subscriptions
Resources
Letta Alternatives
Explore 52 products in the Agents category. View all Letta alternatives.
AutoGen
Microsoft's multi-agent framework, in maintenance mode since 2026; Microsoft points new projects to Agent Framework
Burr
Build stateful AI agents and applications as state machines, with a built-in tracing UI
Superagent
Prototype and deploy agents powered by large language models.
Smolagents
Lightweight Python agent framework from Hugging Face where agents write Python code instead of JSON tool calls
Semantic Kernel
Microsoft's SDK for building and orchestrating AI agents in .NET, Python, and Java
Langroid
Multi-agent LLM framework using message-based task delegation inspired by the Actor model
Work on Letta? Feature it at the top of Agents.
Is your product missing?