Last week, Google quietly released an app that allows users to run a range of open AI models from the Hugging Face development platform locally on their phones.
The app is called Google AI Edge Gallery and is available for Android. An iOS version will follow soon. Users can use the app to search for, download, and run suitable models to generate images, answer questions, or write and edit code, for example. Everything happens locally on the device, without an internet connection. The app uses the computing power of supported phones.
Although AI models in the cloud are often more powerful, there are also disadvantages. Some people prefer not to send personal or sensitive data to external servers, or they want to use AI in places without an internet connection.
App in experimental alpha version
Google calls the AI Edge Gallery an experimental alpha version. The app can be downloaded from GitHub and displays shortcuts to AI functions such as image analysis or chatting on the home screen. Selecting a function brings up a list of suitable models, including Google’s Gemma 3n.
The app also offers a Prompt Lab, where users can start a task with a single input. Examples include summarizing or rewriting text. The Prompt Lab contains templates and adjustable parameters to customize the behavior of the models. Users can choose a language model themselves via Hugging Face integration, or use their own language model via LiteRT.
Google warns that performance may vary. Newer phones with more powerful hardware generally run the models faster. The size of a model also plays a role: larger models take longer to perform a task than smaller ones.
Google invites developers to provide feedback on their experience with AI Edge Gallery. The app is licensed under the Apache 2.0 license and can therefore be used with virtually no restrictions, including in commercial applications.
Also read: Cisco wants to use AI to defend AI at machine scale