Letta

What is Letta?

Letta is an AI research lab and software maker focused on agents that learn from experience. Its product, Letta Agent, is described as a self-improving AI agent whose memory, identity, and capabilities evolve over time. The work builds on the lab's research into memory models, context management, continual learning, and sleep-time compute.

Rather than treating every interaction as an isolated prompt, Letta's approach centers on maintaining and changing agent state. This is intended for developers and organizations building long-running agents that need to retain experience and develop a consistent operating context.

Core Features

• Stateful agent design: Maintains agent memory and identity across experiences.
• Evolving capabilities: Supports agents whose behavior and capabilities can change as they learn.
• Memory-focused research: Applies work on memory models, context repositories, and continual learning to software.
• Developer installation: Letta Agent is distributed through the @letta-ai/letta-code package.

Use Cases

• Long-running assistants: Developers can create agents that retain useful context across repeated work.
• Experiential agents: Research teams can explore systems that improve from accumulated interactions.
• Stateful automation: Organizations can build workflows in which agent identity and memory persist instead of resetting each session.

Frequently Asked Questions

What makes a Letta agent stateful?

Its memory, identity, and capabilities are designed to persist and evolve with experience rather than being limited to a single interaction.

Is Letta only a research organization?

No. The lab publishes research and ships that work as software through Letta Agent.