Letta Code is a stateful agent harness for creating AI agents that behave more like people than tools, with persistent memory, identity and the ability to learn over time. Note that this is a distinct project from the other similarly named product already listed on Tooloso: this entry covers the open-source Letta Code harness repository itself.
Letta Code is developed by the Letta team (creators of MemGPT) on GitHub, where it has about 3,670 stars, with very active development in 2026. Agents built with it continually rewrite their own memory, skills and prompts, and can even modify the harness through mods, so behaviour improves over long horizons instead of resetting every session. You can interact with agents through a local CLI, in the browser at chat.letta.com, or through messaging integrations such as Telegram, Slack and Discord. Features like periodic dreaming with /sleeptime, behaviour inspection with /doctor and cross-agent message search make long-lived, always-on agents practical. Because agents accumulate skills and memories, a Letta Code agent set up for a project keeps getting better at that project’s conventions, reviewers and tools, which is a fundamentally different proposition from a stateless coding assistant that starts cold every morning.
Letta Code is open source under the Apache-2.0 license and free to self-host. Letta also offers a hosted platform and cloud service with paid tiers for teams that prefer not to run the infrastructure themselves.
Developers building long-lived personal or team agents that must remember context, learn preferences and keep working across sessions, rather than stateless chatbots.