LlamaBarn – automatically configure models based on your Mac's hardware (github.com)

LlamaBarn is a tiny (~12 MB) macOS menu-bar app that makes running local LLMs on a Mac as simple as clicking a model from a curated catalog. It automatically configures each model for your Mac’s hardware, starts a local server at http://localhost:2276, and exposes a built‑in web chat UI plus a familiar REST API so apps can call local models just like cloud APIs. The app keeps everything contained in ~/.llamabarn (no system‑wide installs), is written in Swift, and is installable via brew install --cask llamabarn or from Releases. For developers and non‑technical users this lowers the barrier to local inference—no manual tuning or server management required—while preserving privacy and offline use. LlamaBarn builds on llama.cpp/llama-server and supports the same endpoints (e.g., /v1/health and /v1/chat/completions), embedding and completion models, parallel requests, multiple concurrent models, and vision‑capable models where available. Because it auto‑tunes runtime settings for your hardware (so models run stably and efficiently), it’s useful both as a drop‑in local LLM provider for apps and as an easy way for everyday users to experiment with local models.