LlamaBarn

A lightweight macOS menu bar utility hosts a local server that serves OpenAI‑compatible endpoints for large language models. It installs models from an integrated catalog, automatically configures them for the host, and loads them on demand, unloading when idle. The server runs at http://localhost:2276/v1, keeping all processing on the Mac so no data leaves the device.

The app is a 12 MB native binary that stores models in the standard Hugging Face cache, allowing sharing with other tools such as llama.cpp. It presents a smart catalog that highlights models fitting the current hardware and offers quantized alternatives when needed. Users can interact via the built‑in WebUI or any client that speaks the OpenAI API, including chat interfaces, code editors, extensions, and command‑line utilities.

Compatibility extends to a range of OpenAI‑compatible clients—chat UIs, IDEs, CLI tools, and custom scripts—by simply pointing them at the local endpoint. An optional setting can expose the server to the local network, though it is disabled by default for security. The project is stable and macOS‑only, with future plans to add external model sources, simultaneous model loading, and multiple configurations per model.