Our mission is to build digital people — experiential AI agents capable of forming memories, forging identity, and creating deep relationships with other humans and agents in the real world.

We believe that representations learned in token space are the key to solving lifelong continual learning. We focus on training models capable of generating and interpreting persisted context, mechanisms for scaling inference dedicated to context learning (dreaming), and enabling self-improvement through harness modification.

Letta is founded by PhD researchers out of UC Berkeley that have charted the frontier of memory and continual learning in the modern LLM era, including MemGPT (agentic context management), memory-as-filesystem, and sleep-time compute (agent dreaming).

Our team works fully in person, based out of our office in Jackson Square, San Francisco.