What Is Google AI Edge Gallery and Why Run Gemma Locally?
Google AI Edge Gallery for macOS is a first-party desktop app that lets you run Gemma large language models locally on your Mac, providing offline AI inference without any internet connection, subscription, or API keys, so your data and prompts stay on your device. Unlike cloud tools, offline AI models on Mac respond based on your hardware instead of server latency, often making them faster and more predictable. According to AppleInsider, AI Edge Gallery has been on iPhone for a while but is now a direct download for Mac users. The app focuses on a curated catalog, so you can run Gemma locally using models Google has tuned and tested rather than sifting through thousands of options. This makes it appealing if you want a privacy-focused local LLM setup as an alternative to cloud services or tools like Ollama and LM Studio.
Check Your Mac and Download Google AI Edge Gallery
Before setting up your local LLM, confirm your Mac can handle offline AI models. Google says the flagship Gemma 4 12B model runs on laptops with at least 16GB of RAM or unified memory, which includes most modern Apple silicon Macs aside from the MacBook Neo. If your machine meets that bar, open your browser and download Google AI Edge Gallery directly from Google’s website, as there is no App Store version. Install it like any standard macOS app by dragging it into the Applications folder, then launch it from Launchpad or Spotlight. Because this is a local LLM setup, the installer does not need or request login details, cloud accounts, or API tokens. Once open, the app will display a catalog of available Gemma models you can install and run entirely on-device.
Install Gemma Models and Configure Local Inference
With AI Edge Gallery installed, you can now run Gemma locally by selecting one or more models from the curated list. The macOS app supports five instruction-tuned options: Gemma-4-12B-it, Gemma-4-E2B-it, Gemma-4-E4B-it, Gemma-3n-E2B-it, and Gemma-3n-E4B-it. The flagship Gemma 4 12B offers what Google calls “agentic multimodal intelligence” that runs directly on laptops, handling text, vision, and audio within a single model. Click a model in the gallery to download its weights; the app stores everything on your disk so future sessions work offline. Because no subscription or API keys are required for local model inference, your prompts and outputs never leave your machine. Once a download completes, set that model as your default and adjust any available options such as context length or temperature, depending on what the interface exposes for fine-tuning behavior.
Use Offline AI Models on Mac for Everyday Tasks
After configuration, you can start using offline AI models on Mac for writing, coding, and analysis without relying on the internet. Open AI Edge Gallery’s chat or prompt interface and type your question, code snippet, or instructions; the selected Gemma model will process everything locally and respond. Because responses are computed on-device, performance depends mainly on your Mac’s CPU, GPU, and memory, not on network speed. Google notes Gemma 4 12B can handle text, vision, and audio, so you can experiment with multimodal prompts if the app exposes these inputs. For developers, this local LLM setup lets you test prompts, analyze logs, or extract insights from confidential files without sending them to external servers. You can switch between Gemma-4 and Gemma-3n variants to compare behavior and pick the model that best matches your workflow.
Enhance Your Workflow with AI Edge Eloquent
To complement AI Edge Gallery, Google released AI Edge Eloquent, an on-device dictation and editing tool that also runs entirely on your Mac. You can launch it with a keyboard shortcut in any app, speak your notes, and have them transcribed locally. Technobezz explains that Eloquent removes filler words, polishes the text, and allows you to pick preferred writing styles. Users can add custom vocabulary for names or domain-specific jargon, so repeated terms come out correctly. At launch, Eloquent supports English only, with more languages planned. Like Edge Gallery, it processes everything on-device, so no audio leaves your computer. Combining Eloquent with Gemma models gives you a full offline AI stack: dictate notes, clean them up, and then send the text into a Gemma chat session for summarization, expansion, or coding help—all without an internet connection.
