Letta
Open-source framework for building stateful AI agents with long-term memory, self-editing capabilities, and unlimited context windows.
13,000+
GitHub Stars
2023
Founded
10K+
Developers
Overview
Letta (formerly MemGPT) is an open-source framework for building stateful AI agents with unlimited context through intelligent memory management. It enables agents to remember and learn from all past interactions by implementing a virtual context management system inspired by operating system memory hierarchies. Agents can self-edit their memory, reason over conversation history, and maintain persistent state across sessions—unlocking long-running agents that improve over time.
The Verdict
Who Should Use Letta?
Best For
- Developers building conversational agents requiring long-term memory
- Applications needing agents that learn and improve from user interactions
- Multi-session chatbots requiring context retention across conversations
- Research teams exploring stateful agent architectures
- Companies wanting self-hosted, open-source memory management
Not Ideal For
- Simple stateless chatbot use cases without memory requirements
- Teams seeking fully managed cloud memory services
- Production deployments requiring enterprise SLA and support
What's Great
- Virtual context management breaks through LLM context window limitations
- Self-editing memory allows agents to summarize and organize their own knowledge
- Open-source with Apache 2.0 license—full control and customization
- Model-agnostic design works with OpenAI, Anthropic, open-source LLMs
- Python SDK with simple API for agent creation and memory management
- Active research-backed development from UC Berkeley team
Watch Out For
- Early-stage project with evolving APIs and breaking changes
- Memory management adds latency compared to stateless agents
- Requires understanding of memory architecture for optimal performance
- Limited production deployment examples and enterprise features
Pricing
Open Source
Free
Apache 2.0 licensed. Use with your own LLM API keys. Self-hosted with no usage limits or restrictions.
Letta Cloud (Beta)
Contact
Managed hosting with built-in LLM access. Currently in private beta—contact for early access and pricing.
View all features & details
Key Features
- Virtual context management system
- Self-editing agent memory
- Hierarchical memory storage (core, archival, recall)
- Multi-model support (OpenAI, Anthropic, local models)
- Python SDK and REST API
- Persistent agent state across sessions
Platforms
- Python (primary)
- Linux/macOS/Windows
- Docker containers
- Cloud deployment ready
How It Compares
| Feature | Letta | Zep | LangChain Memory |
|---|---|---|---|
| Approach | Virtual context OS | Graph-based memory | Simple buffer/vector |
| Self-editing | Yes, agents edit memory | No, managed externally | No |
| Open Source | Fully open (Apache 2.0) | Open-core model | Fully open (MIT) |
| Memory Architecture | Hierarchical (core/archival) | Knowledge graphs | Key-value stores |
| Best For | Unlimited context agents | Production RAG + memory | Quick LangChain integration |
User Reviews
Loading reviews...