Letta

Letta is an AI research lab based in San Francisco focused on building stateful AI agents with persistent memory. Its work centres on enabling artificial intelligence systems that can continuously learn and improve from their own experience, a capability often referred to as continual learning. The company's technical domains span AI agents, persistent memory, memory management, and large language models.

The company's main product, Letta Code, is a model-agnostic agent framework. It provides language models with persistent memory, computer-use capabilities, skills, sub-agents, and various deployment paths. The tool is designed to work with any underlying large language model rather than being tied to a specific provider.

Letta emphasises transparency in how its agents operate. The company allows developers to inspect agent behaviour at the token level, giving them visibility into the reasoning and memory processes that drive agent actions. This approach is paired with a focus on user ownership of agent data and behaviour.