Letta (previously MemGPT)

Homepage // Documentation // Letta Cloud

👾 Letta is an open source framework for building stateful LLM applications. You can use Letta to build stateful agents with advanced reasoning capabilities and transparent long-term memory. The Letta framework is white box and model-agnostic.

Note

Looking for MemGPT? You're in the right place!

The MemGPT package and Docker image have been renamed to letta to clarify the distinction between MemGPT agents and the API server / runtime that runs LLM agents as services.

You use the Letta framework to create MemGPT agents. Read more about the relationship between MemGPT and Letta here.

⚡ Quickstart

The two main ways to install Letta are through pypi (pip) or via Docker:

• pip (guide below) - the easiest way to try Letta, will default to using SQLite and ChromaDB for the database backends
• Docker (guide here) - recommended for production settings, will default to using Postgres (+ pgvector) for the database backend

Step 1 - Install Letta using pip

$ pip install -U letta

Step 2 - Set your environment variables for your chosen LLM / embedding providers

$ export OPENAI_API_KEY=sk-...

For Ollama (see our full documentation for examples of how to set up various providers):

$ export OLLAMA_BASE_URL=http://localhost:11434

Step 3 - Run the Letta CLI

You can create agents and chat with them via the Letta CLI tool (letta run):

$ letta run

🧬 Creating new agent...
? Select LLM model: letta-free [type=openai] [ip=https://inference.memgpt.ai]
? Select embedding model: letta-free [type=hugging-face] [ip=https://embeddings.memgpt.ai]
->  🤖 Using persona profile: 'sam_pov'
->  🧑 Using human profile: 'basic'
->  🛠️  8 tools: send_message, pause_heartbeats, conversation_search, conversation_search_date, archival_memory_insert, archival_memory_search, core_memory_append, core_memory_replace

🎉 Created new agent 'InspiringSpinach'

Hit enter to begin (will request first Letta message)

💭 User engaged. Recognize as Chad. Acknowledge presence and establish connection. Explore interests and engage in meaningful discourse.
🤖 Hello, Chad. It's nice to meet you. I am keen on learning and exploring new ideas and concepts. Is there a specific topic you'd like to discuss today?

> Enter your message: my name is Brad, not Chad...

💭 A discrepancy in identity. Reflect this change in core memory. Apologize for the oversight, and rectify the error.
⚡🧠 [function] updating memory with core_memory_replace
         First name: Chad
        → First name: Brad
💭 Apologize for the mistake in recognizing his name, and begin the conversation anew, addressing him by the correct name this time.
🤖 I apologize for the oversight, Brad. Let's restart. It's a pleasure to meet you, Brad. Is there a specific topic you'd like to discuss today?

> Enter your message:

Step 4 - Run the Letta server

You can start the Letta API server with letta server (see the full API reference here):

$ letta server

Initializing database...
Running: uvicorn server:app --host localhost --port 8283
INFO:     Started server process [47750]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://localhost:8283 (Press CTRL+C to quit)

When you start the Letta API server, the ADE (Agent Development Environment) will be available on http://localhost:8283:

In Letta, all agents are stored/persisted in the same database, so the agents you create in the CLI are accessible via the API and ADE, and vice versa. Check out the quickstart guide on our docs for a tutorial where you create an agent in the Letta CLI and message the same agent via the Letta API.

🤗 How to contribute

Letta is an open source project built by over a hundred contributors. There are many ways to get involved in the Letta OSS project!

• Contribute to the project: Interested in contributing? Start by reading our Contribution Guidelines.
• Ask a question: Join our community on Discord and direct your questions to the #support channel.
• Report ssues or suggest features: Have an issue or a feature request? Please submit them through our GitHub Issues page.
• Explore the roadmap: Curious about future developments? View and comment on our project roadmap.
• Join community events: Stay updated with the event calendar or follow our Twitter account.

---

Legal notices: By using Letta and related Letta services (such as the Letta endpoint or hosted service), you are agreeing to our privacy policy and terms of service.