This project makes Letta agents available through an OpenAI-compatible API. Agents are listed as models, and sending messages through the chat completion API will send messages to the selected Letta agent and receive reasoning messages.
This project uses uv to run the application. The usual uv methods apply:
uv sync
uv venv
source .venv/bin/activate
You will need a letta server to run. The easiest way to do this is to go to https://github.com/wsargent/groundedllm -- set up the tokens and run docker compose up to bring up the system.
Copy the env_example to .env and set up your credentials:
LETTA_BASE_URL=http://your-letta-server
LETTA_API_TOKEN=your-letta-password-if-any
The server uses Hayhooks to run:
uv run python app.py
The server will come up at http://localhost:1416
Please see the Hayhooks documentation for logging and configuration options.
You can use any OpenAI API compatible client to chat with the agent. I prefer Open WebUI but there are many options.
For your convenience, a simple command line client is included that you can run standalone:
uv run python cli_client.py
And then type /models to list the available models.
Unfortunately, Apple's Terminal app doesn't support clickable hyperlinks: I recommend you use iTerm2 or another terminal that supports OSC8 as it makes clicking on links much easier.
You cannot use tools or upload data sources with an agent currently.