Google Quietly Launches App to Run AI Models Locally

Google has quietly introduced a new app that enables users to run AI models directly on their smartphones without needing an internet connection. The app, called Google AI Edge Gallery, is currently available for Android devices, with an iOS version expected soon.

The app allows users to browse, download, and run a variety of AI models sourced from Hugging Face, a popular platform for open-access AI models. Capabilities include generating images, answering questions, writing and editing code, and more — all processed locally on supported devices without relying on cloud servers.

While cloud-based AI models often offer greater processing power, local deployment offers clear advantages for privacy-conscious users and those needing offline access. By running models directly on the phone’s hardware, users avoid sending sensitive data over the internet.

Labeled as an “experimental Alpha release,” Google AI Edge Gallery can be downloaded from GitHub following specific installation instructions. The app’s home screen features shortcuts to various AI tasks such as “Ask Image” and “AI Chat.” Users can browse compatible models like Google’s Gemma 3n for different tasks.

The app also includes a Prompt Lab, where users can engage in "single-turn" tasks such as summarizing or rewriting text, using customizable templates and settings to fine-tune model behavior.

Performance may vary based on hardware and model size, Google notes. High-end smartphones will naturally handle models faster, while larger models may require more time to complete tasks.

Licensed under Apache 2.0, Google AI Edge Gallery is open for commercial and non-commercial use without significant restrictions. Google is actively seeking feedback from the developer community to refine the app as it evolves.