Letta is an open-source platform for stateful AI agents, agents with long-term memory that they can edit themselves, which grew out of the MemGPT research project.
In plain words
A language model forgets everything outside its context window. MemGPT, a research project from UC Berkeley, borrowed an idea from operating systems: move information between fast and slow memory so the model seems to have more. Letta builds on that. It saves all of an agent's state in a database, puts key memories, organised as blocks, into the prompt, and gives the agent tools to rewrite its own memory. The current code lives in the letta-code repository.
A quick example
Letta's docs show an agent created with a memory block labelled "human" that describes the user. When the user later chats, the agent draws on that block to recognise who it is talking to, and it can update the block if it learns something new.
Where you'll meet it
- Personal assistants that remember users.
- Coding agents that learn over time.
- Apps that embed long-running stateful agents.
Related
Sources
- PaperMemGPT: Towards LLMs as Operating Systems, Packer et al. (arXiv) · read 29 Sept 2026
- DocsMemory, Letta Docs · read 29 Sept 2026
- Docsletta-ai/letta, Letta (GitHub) · read 29 Sept 2026
- DocsMemGPT Is Now Part of Letta, Letta · read 29 Sept 2026
- DocsLetta Documentation, Letta Docs · read 29 Sept 2026
- DocsMemGPT concepts, Letta Docs · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 11 Nov 2026. Spotted an error? Report it.