Letta is an open-source framework for building stateful AI agents with advanced memory, reasoning, and context management. It provides a white-box, model-agnostic architecture, allowing developers to create and deploy agents that can maintain long-term memory and interact with various Large Language Models (LLMs).

How It Works

Letta agents are managed within a persistent Letta server, which stores agent data in a database (PostgreSQL recommended, defaults to SQLite for pip installs). The server exposes a REST API and SDKs for Python and TypeScript, enabling interaction with agents. A graphical Agent Development Environment (ADE) is provided for managing, testing, and observing agents. The framework supports multiple LLM backends (OpenAI, Anthropic, vLLM, Ollama) and offers tools for memory management, conversation search, and more.

Quick Start & Requirements

• Docker: docker run -v ~/.letta/.persist/pgdata:/var/lib/postgresql/data -p 8283:8283 -e OPENAI_API_KEY="your_openai_api_key" letta/letta:latest
• Pip: pip install -U letta followed by export OPENAI_API_KEY=sk-... and letta run for CLI or letta server for API.
• Prerequisites: LLM API keys (e.g., OpenAI), Docker (recommended for PostgreSQL persistence and migrations).
• Resources: Requires an LLM provider. Docker setup is straightforward. Pip install defaults to SQLite, with PostgreSQL support via LETTA_PG_URI.
• Links: Homepage, Documentation, ADE, Letta Cloud.

Highlighted Details

• Model-agnostic architecture supporting various LLM providers.
• Persistent agent storage with a dedicated server and REST API.
• Graphical Agent Development Environment (ADE) for agent management.
• CLI tool for direct agent interaction and creation.

Maintenance & Community

• Active open-source project with over a hundred contributors.
• Community support via Discord.
• Roadmap and contribution guidelines available.

Licensing & Compatibility

• Licensed under Apache 2.0.
• Permissive license suitable for commercial use and integration into closed-source applications.

Limitations & Caveats

Database migrations are not officially supported for the SQLite backend used by default with pip installs; PostgreSQL is recommended for robust data migration across versions. The project was formerly known as MemGPT.