Letta logo

Letta

Free tier

An AI research lab building self-improving agents that remember everything and learn continuously.

Free tier available·All audiences·API available·Open source

Key strengths

Persistent, evolving agent memory and identityContinual learning — agents improve over time without retrainingGit-based memory versioning for reproducible context managementSleep-time compute for offline reasoning during idle periodsResearch-backed foundations from the creators of MemGPT at UC Berkeley
Free tier + paid plans
San Francisco, USA
Founded 2023
Self-hostable
No ratings yet

Developer Docs

Getting started with Letta is straightforward via the npm CLI:

npm i -g @letta-ai/letta-code

Core Concepts:

  • Context Constitution — A principled framework governing how agents manage and update their context to enable learning from experience.
  • Context Repositories — Git-based memory storage that brings programmatic context management and version control to agent memory, enabling reproducibility and rollback.
  • Continual Learning in Token Space — The underlying mechanism by which Letta agents improve over time by learning directly in the token representation space, without requiring full model retraining.
  • Sleep-time Compute — Agents can perform offline reasoning and memory consolidation during idle periods, improving future response quality without added inference latency.

Integration & Deployment:

  • Letta Agent exposes APIs for embedding agents into your application stack.
  • Agents maintain persistent, versioned memory stores that evolve with usage.
  • Designed for long-horizon tasks where stateful context across many sessions is critical.
  • Community support available via GitHub and Discord.