Local AI models in your Windows system tray, powered by llama.cpp
Llama is a free, open-source tray app for Windows 11 that runs open AI models entirely on your own machine. It comes from the llama.cpp team together with Hugging Face, and its whole job is to make local AI a one-click affair: no subscription, no account, no prompt leaving your computer. It is MIT licensed, and the current release is v0.14.1, published on October 30, 2026.
The app manages llama.cpp for you: it uses your existing llama.exe or downloads one, then serves models over a local API your other apps can talk to. Browsing and downloading recommended models from the Hugging Face Hub takes one click, and each model shows context-length options with per-option memory estimates read from the GGUF header. Pressing Alt+Space from any app opens a Spotlight-like chat overlay that hides instead of closing, so your conversation is still there when you come back.
The catches are the usual ones for local AI: models are large downloads and need real disk space and memory, and the app itself is young at version 0.x. It is Windows 11 only (Mac users get the sibling app it was ported from). What you get in return is rare: a polished, private, genuinely free way to run open models, with zero telemetry by design.
Featured in our discovery: Llama: Your Own Private AI in the Windows Tray
A borderless, Mica-backed flyout anchored to the tray icon puts your local models one click away.
Browse and download recommended models from the Hugging Face Hub, with per-context-length memory estimates read from the GGUF header.
Uses your existing llama.exe or downloads one for you, then manages the local server and its REST API.
A Spotlight-like window opens from any app; it hides instead of closing, so your conversation stays.
No. Inference is fully local. Per the official README, the only outbound traffic is model downloads, release-update checks and calls to the local server, and the app sends no telemetry at all.
Windows 11 on x64 or ARM64. The app handles the llama.cpp setup itself and downloads models from the Hugging Face Hub in one click.