Letta is an AI research lab and software company in San Francisco that develops stateful AI agents. A stateful agent preserves information across interactions instead of treating every request as an isolated exchange. Letta's work focuses on agents that can retain memory, learn from experience, use computers, and continue operating across models and interfaces.
Research lineage
Letta was founded by the researchers behind MemGPT, an architecture that treated a language model's context window as one tier in a larger memory hierarchy. Later research broadened that question from moving information between memory tiers to the structures that let agents consolidate experience, compile context, and improve over time.
Software
Letta's public software is organized around several related systems:
- Letta Code is the open, model-agnostic runtime for persistent agents, computer use, skills, subagents, and deployment.
- Letta Agent is the personalized stateful agent built on that open harness and memory research.
- The Letta SDK exposes agent primitives for developers building applications and services.
The organization describes Letta Code as its flagship runtime. Its current architecture places agent memory in git-backed files, packages reusable capabilities as skills, and delegates work through subagents rather than binding those functions to one model provider.