Letta Code is an open, model-agnostic runtime for stateful AI agents developed by Letta. It surrounds a language model with persistent memory, computer access, reusable skills, subagents, scheduling, permissions, and deployment paths. Because those capabilities live in the harness rather than one provider's model, an agent can change models without discarding its surrounding state.
Memory-first runtime
Letta Code stores agent context in a git-backed memory filesystem called MemFS. The agent can inspect and revise its own memory as files, while version history preserves how that context changed. This extends the memory-management lineage of MemGPT from specialized memory tools toward general filesystem operations and versioned context repositories.
Extension and execution
Skills package reusable procedures and domain knowledge. Subagents divide work into separate contexts. Hooks and mods extend runtime behavior, permissions constrain actions, and schedules allow work to recur or continue later. Local, desktop, web, and messaging interfaces can route interaction to the same persistent agent.
Letta Agent is the personalized agent product built through this runtime. Letta Code is also a developer tool in its own right: it can be used directly for coding, research, operations, and other computer-based work.