# Llama app: free offline AI for Mac and Windows (November 2026)

What the free Llama desktop app from the llama.cpp team and Hugging Face does, which models it runs, how much memory you need and how to install it.

URL: https://aitoolsthisweek.com/llama-app
Published: 2026-11-14T03:27:40.863Z
Updated: 2026-11-14T03:28:01.461Z

**Llama is a free app that runs AI chatbots on your own computer, with no account, no subscription and no internet once a model is downloaded.** It comes from the team behind llama.cpp, the engine inside many local AI tools, together with Hugging Face, and it showed up this week next to the long-awaited **llama.cpp v1.0.0** release on **November 13, 2026**.

If you've wanted a private ChatGPT-style assistant that never sends your words to a server, this is the simplest official way to try it so far.

## What the Llama app does

Llama lives in your **Mac menu bar** or **Windows 11 system tray**. From there you open a chat window that looks and works like a normal AI chatbot, except the model runs on your own machine. According to [llama.app](https://llama.app):

- **It works offline.** After a model is downloaded, nothing needs the internet.
- **It picks settings for you.** Llama chooses how much to compress the model, how much text it can remember and other technical options for your hardware. Experts can still edit everything in a plain-text file.
- **It recommends models for your computer.** You get a "recommended for this Mac/PC" list and install with one click.
- **It saves memory when idle.** Models load when you use them and unload after **5 minutes** of no use.
- **Other apps can use it.** It runs a local server that speaks the same "language" as the OpenAI and Anthropic APIs, so coding tools and chat apps can connect without an API key or a bill.
- **You can reach it from your phone** or another device over Tailscale.

The Mac app is tiny: about **1 MB to download** and **4 MB installed**. The models are the big part (several gigabytes each).

## Which computer do you need?

The site lists memory (RAM) guidance for its starter models:

| Model | Suggested memory | Good for |
|---|---|---|
| Gemma 4 (Google) | 16 GB or more | Everyday chat, writing help |
| gpt-oss (OpenAI) | 24 GB or more | Stronger reasoning |
| Qwen 3.8 (Alibaba) | 32 GB or more | The most capable of the three |

llama.app also names smaller options such as **Gemma 4 E4B** and **gpt-oss 20B**. The site doesn't publish a minimum macOS version or chip, so let the app's "recommended for this Mac/PC" list guide you: it only suggests models your machine can run.

Not sure whether local or cloud AI suits you? Our guide to [on-device vs cloud AI](/on-device-vs-cloud-ai) explains the trade-offs, and [AI models on your phone](/ai-models-on-your-phone) covers the mobile side.

## How to install Llama

1. Go to **llama.app** and choose **Download for Mac** or **Download for Windows 11**.
2. Mac users who already use Homebrew can type `brew install --cask llama-app` in Terminal instead.
3. Open the app. It appears as a small icon in the menu bar (Mac) or system tray (Windows).
4. Pick a model from the recommended list and wait for the download.
5. Open the chat window and start typing.

On **Linux** there's no desktop app yet, only a command-line install from the same site.

If you already use llama.cpp, Llama finds your existing models automatically, because both store them in the same Hugging Face folder.

## Is it really free, and what's the catch?

The app is **free and open source**; the Mac version's code is on GitHub under the **MIT license**. There's nothing to pay because there's no server: your computer does the work. The catches are practical, not financial:

- **Local models are smaller than ChatGPT, Claude or Gemini.** They're good for drafting, summarising and questions, but they won't match the top cloud models on hard tasks.
- **Speed depends on your hardware.** A recent Mac with plenty of memory feels quick; an older laptop will be slow.
- **No web search built in.** A local model only knows what it learned in training.

For free cloud options with no install, see our list of [free AI tools](/free-ai-tools).

## What llama.cpp 1.0 means

The Hugging Face announcement calls v1.0.0 "a stronger foundation." In plain terms: after about **3.5 years**, llama.cpp now follows clear version numbers, so apps built on it (Llama, and many others) are less likely to break after updates. It supports **more than 150** model architecture variants and ships through Homebrew, Winget, Docker and other installers on Windows, Mac, Linux, Android and iOS.

*AI Tools This Week is not affiliated with Hugging Face, ggml-org or Meta.*

## Frequently asked questions

### Is the Llama app free?

Yes. llama.app calls it free and open source, and the Mac version's code is on GitHub under the MIT license. There is no subscription and no usage bill, because the models run on your own computer.

### Does Llama work without the internet?

Yes, once a model is downloaded. llama.app says it works offline, and the Mac app's README says models run on your Mac and no data leaves it. You need a connection only to download the app and models.

### Is Llama the same as Meta's Llama models?

No. This Llama is an app made by the llama.cpp team (the ggml-org group) and Hugging Face. It runs many open models, such as Gemma 4, gpt-oss and Qwen 3.8; the name comes from llama.cpp, the engine underneath.

## Sources

- [Llama — Your AI. On your computer. (llama.app)](https://llama.app)
- [llama.cpp v1.0.0: A stronger foundation for the local AI ecosystem (Hugging Face blog, Nov 13, 2026)](https://huggingface.co/blog/ggml-org/llama-cpp-v1)
- [Llama-macOS README (ggml-org on GitHub)](https://github.com/ggml-org/Llama-macOS)
