# Letta

> Research lab behind MemGPT building stateful agents with persistent, self-improving memory

Source: https://dontaskagain.com/letta
Website: https://www.letta.com/?utm_source=dontaskagain
Pricing: Freemium
Categories: [Memory](https://dontaskagain.com/categories/memory.md), [Agent frameworks](https://dontaskagain.com/categories/agent-frameworks.md), [Open source](https://dontaskagain.com/categories/open-source.md)
Listed: 2026-10-18
Updated: 2026-10-18

Letta is an AI research lab in San Francisco, founded by the creators of MemGPT at UC Berkeley, that builds agents with long-term memory. Its research ships as software, mainly Letta Agent, a self-improving agent whose memory, identity and capabilities evolve with experience, and the Letta Code command-line tool.

**Key features:**

- Give agents persistent memory that carries across sessions.
- Manage context with programmatic context management and Git-based context repositories.
- Let agents learn continuously in token space instead of through retraining.
- Use memory models trained with memory-native reinforcement learning.
- Benchmark continual learning with FrontierMemory.
- Install Letta Code globally from npm.

Letta grew out of the MemGPT paper on virtual context management and lists case studies with Bilt, 11x, Kognitos and Hunt Club. Its agent software is open source; the homepage does not show pricing or describe a hosted plan.

## More memory agents

- [Mem0](https://dontaskagain.com/mem0.md): Open-source memory layer that lets AI agents store, compress and recall user context across sessions
- [Zep](https://dontaskagain.com/zep.md): Context layer that builds governed, time-aware context graphs from business data so agents retrieve what tasks need
- [Supermemory](https://dontaskagain.com/supermemory.md): Memory and continual-learning engine that stores agent context in a time-aware graph and injects it at runtime