Hugging Face has officially announced Llama Windows, a free, open source app that lets people run open AI models on their own Windows 11 PC without touching a command line. “Today we’re announcing Llama Windows, a free and open source app that puts llama.cpp in your Windows 11 tray,” the company wrote in a blog post on 13 October 2026 [1].
The announcement is new, but the app itself is not brand new. Its code repository on GitHub, ggml-org/Llama-Windows, lists public releases going back to an early alpha version on 27 July 2026, with the latest, version 0.12.0, published on 10 October [2].
What llama.cpp is, and what “running a model locally” means
Most people use AI chatbots through a website or app that sends each question to a company’s computers in a data centre. Running a model locally means the AI model, a large file of learned numbers, sits on your own computer, and your own processor and graphics chip do the work. Your questions do not need to leave the machine.
llama.cpp is the free software that does that work. It loads an AI model and generates answers on ordinary computers, from laptops to desktops. Hugging Face, which runs the biggest public library of open AI models, calls it “the most widely used engine for running open models locally” [1]. It also introduced a file format called GGUF, which Hugging Face says “is the standard for local models on the Hugging Face Hub” [1]. GGUF files are usually “quantized”: the model’s numbers are stored with less precision so the file is smaller and fits on a normal PC, at some cost in quality.
Until now, Hugging Face says, using llama.cpp on Windows “still meant a terminal”: picking the right version for your hardware, finding a model, choosing between quantized builds and starting a server with the right settings [1].
What Llama Windows does
The app turns those steps into a few clicks. “Pick a model from the Hugging Face Hub, download it in one click, and run it locally from the notification area. No terminal, no build step, no admin rights,” Hugging Face writes [1].
According to the blog post, Llama Windows:
- Lives as a llama icon in the Windows notification area (the system tray). Clicking it opens a small panel; there is no main window to manage.
- Shows a list of recommended, ready-to-run models, each with its parameter count (a rough measure of model size), its size on disk and whether it can understand images, so you can pick one that fits your PC before downloading.
- Lists the models already on your PC, lets you load or unload them, and opens llama.cpp’s own web chat page.
- Installs llama.cpp itself using the official installer if it is not already there, without needing administrator rights.
- Offers a pop-up chat window you can open from anywhere with Alt+Space.
Hugging Face calls it “the Windows counterpart to llama.app on macOS” [1].
Privacy
Hugging Face pitches the app as private by design. In the company’s words: “Everything done through Llama Windows stay on your machine, no internet connection needed, no data saved elsewhere” [1]. That is Hugging Face’s own claim about its software. You do still need an internet connection to download a model in the first place.
Who it is for, and what you need
Hugging Face says the app is aimed at “Local AI Builders” who want llama.cpp’s performance in a native Windows app, and that it is also “the easiest way for any Windows user to run an open model on their own PC for the first time” [1].
For developers, the app runs llama.cpp as a local server at http://localhost:9931, with an OpenAI-compatible interface, so existing scripts, apps and agents written for OpenAI-style tools can be pointed at the model on your own PC instead [1]. Models are stored in the standard Hugging Face cache folder, so other Hugging Face tools can reuse them, and users can add a Hugging Face access token to download gated or private models [1].
What you need, according to Hugging Face, is a Windows 11 PC. You download the app from llama.app, launch it, click the tray icon, pick a recommended model, download it, load it and start chatting [1]. Hugging Face does not list minimum memory or graphics requirements; larger models need more memory and disk space, which is why the app shows each model’s size up front.
Windows on Arm and NVIDIA RTX Spark
Hugging Face says Llama Windows “launches with native support for Windows on Arm and for the new NVIDIA RTX Spark” [1]. On RTX Spark machines, it says, llama.cpp runs models on the RTX Spark graphics chip, and “We worked with NVIDIA to have Llama Windows ready for RTX Spark on day one” [1].
Open source
The code is public on GitHub under ggml-org, the organisation behind llama.cpp, and is released under the MIT licence [2]. Hugging Face describes it as a WinUI 3 app plus .NET libraries for the llama.cpp server, the Hugging Face Hub client and the model catalogue, and invites people to file issues and send code [1].
What this means
- Local AI gets a one-click front door on Windows. The hard parts of running llama.cpp, from picking files to starting a server, are hidden behind a tray icon.
- The privacy promise is Hugging Face’s. The app is designed to keep everything on your PC, and its code is open for anyone to check.
- It is an official launch, not a first release. Builds have been on GitHub since July; 13 October is when Hugging Face formally announced it.
Sources
- Hugging Face (Morgan Funtowicz, Jeff Boudier, Emanuil Rusev), “Introducing Llama Windows: llama.cpp in your Windows tray”, 13 October 2026. https://huggingface.co/blog/huggingface/llama-windows
- GitHub, ggml-org/Llama-Windows, repository and releases (latest v0.12.0, 10 October 2026). https://github.com/ggml-org/Llama-Windows/releases
