Letta (formerly MemGPT) is an open-source platform and developer ecosystem designed to build stateful AI agents — systems that keep persistent memory across sessions, learn from interactions, and improve their behavior over time.
Key features and concepts:
- Memory-first architecture: Letta organizes agent state as memory blocks (labeled units of stored context such as persona, user profile, conversation history, facts). These memory blocks allow agents to recall and update long-term information across sessions.
- Multi-provider model support: Letta is model-agnostic and can work with various model providers and embeddings (examples in documentation include OpenAI, Anthropic, Google Gemini), so developers can pick providers or switch between them.
- SDKs & API: Official client libraries for TypeScript/Node.js and Python (npm package @letta-ai/letta-client and pip package letta-client) plus a REST API allow easy integration into applications and services.
- Tools & integrations: Built-in tool support (examples: web_search, run_code) lets agents consult external tools during reasoning. Letta Code provides a memory-first coding harness to interact with persisted agents from the CLI.
- Deployment options: Letta offers a hosted developer platform and documentation for self-hosting a Letta server, enabling privacy-sensitive or on-premises deployments.
- Developer ecosystem: Extensive docs (quickstart, API reference, tutorials), an active OSS repo with many contributors, community channels (Discord, forum, social links) and examples to accelerate adoption.
Typical use cases:
- Personal assistants and chatbots that must remember user preferences, past interactions, and long-term context.
- Autonomous agents that learn from feedback and improve workflows over time.
- Applications that combine multiple tools and external data sources while preserving state across conversations.
Getting started (high-level):
- Install the client: npm install @letta-ai/letta-client or pip install letta-client.
- Create an agent via the Letta API specifying a model, embedding, memory_blocks, and tools.
- Send messages to the agent and let it use memory blocks and tools to maintain state.
- Optionally self-host Letta for full control or use the Letta hosted platform for faster setup.
Licensing & community: Letta is an open-source project maintained by the letta-ai organization on GitHub, with community contributions, Discord/forum support, and public docs. The project emphasizes memory-first agent design to enable persistent, improvable AI behaviors.
(Technical note: repository metadata shows the project was created in October 2023 and includes examples for both TypeScript and Python clients, plus references to Letta Code and documentation at docs.letta.com.)
