Local LLMs from your macOS menu bar — no terminal tab required.
Runs a patched mlx_lm.server against any model in your models folder (~/.llmtray/models by default, or point it at LM Studio's).
Search mlx-format models, download with a real progress bar — speed, ETA, pause/resume, and a size check per file.
Streamed reasoning shown separately from the final answer, tok/s, adjustable temperature/top-p/max tokens.
Green + pulsing while anything is generating — including an external tool hitting the server directly — orange/red under real thermal pressure.
Native Swift/SwiftUI, no Electron, no bundled runtime. The whole app is a ~1MB download and sits idle at a few MB of memory until you actually start a server.
For contributing, or if you'd rather not run an unsigned binary:
git clone https://github.com/ipsupport-llc/llmtray.git
cd llmtray
swift build
.build/debug/LLMTray