Letta is an AI research lab based in San Francisco, dedicated to building machines that can learn. Its vision is to create a future where humans and digital beings coexist: experiential agents that can remember everything, continuously learn, and improve themselves over time.
Letta drives agent learning through memory models trained with memory-native reinforcement learning, and has established a set of principles to govern how AI agents manage context so they can learn from experience. Letta has rebuilt the way agent memory works, using programmatic context management and git-based version control. Learning in token space is key to building AI agents that can truly improve over time. Letta allows agents to reason about context during idle time rather than at inference time. This is the original work on virtual context management—Letta's starting point.
Letta's research is delivered as software: Letta Agent, a self-improving AI agent whose memory, identity, and capabilities evolve with experience. The installation command is: npm i -g @letta-ai/letta-code.
Featured case studies include: Bilt, 11x, Kognitos, Hunt Club.
Media coverage: WIRED's "Do Large Language Models Dream of AI Agents?", Fast Company's "Why sleep-time compute is AI's next big leap", Matthew Berman's "Sleep-Time Compute—AI That 'Thinks' Around the Clock", TechCrunch's "Letta, one of UC Berkeley's most anticipated AI startups, just emerged from stealth."
Letta was founded by the creators of MemGPT at the UC Berkeley Sky Computing Lab, which is also the birthplace of Spark and Ray. Advisors include Ion Stoica and Joey Gonzalez. Supporters include Jeff Dean, Clem Delangue, Cris Valenzuela, Jordan Tigani, Tristan Handy, Barry McCardel, Robert Nishihara, Sunflower Capital, Essence VC, and Felicis Ventures.
Letta is a small team of researchers and engineers dedicated to building machines that can learn.
