Llama MacOS Desktop Controller

Llama MacOS Desktop Controller is a self-hosted desktop application that lets you control macOS with natural language. You type or speak a request like opening an app or arranging windows, and the tool asks a local Llama model to write Python that carries out the action. A React front end handles the input, and a Flask backend talks to LlamaStack, which drives the code generation.

The project was published on GitHub as a demo of LlamaStack, Meta's open stack for running Llama models. It targets developers and tinkerers who want to see how far a small local model can go at turning everyday instructions into working system calls. Because everything runs on the machine, there is no cloud round trip and no account to set up.

The catch is that this is a demo, not a finished product. It runs only on macOS, needs Python, Node.js, Ollama, and a LlamaStack container running before it works, and its safety checks are thin, so a flawed line of generated code can affect the whole system.

Getting Started

1. Install the prerequisites: Node.js 16 or later with npm, Python 3.8 or later with pip, and Ollama, then pull the Llama 3.2 3B Instruct model.
2. Start LlamaStack on port 5001 with Docker, pointing it at your local Ollama server.
3. In the backend folder, install the Python requirements and run server.py so the Flask API listens on port 5066.
4. In the project root, run npm install and npm run dev to launch the React front end on port 5173.
5. Open the browser at localhost:5173, confirm the status indicator turns green, then type or dictate a command and watch the generated code run.