Google has released Google AI Edge Gallery, an Android app that lets people find, download, and run compatible AI models directly on their phones. The app draws from openly available models on Hugging Face and is designed for local use, meaning supported phones can run the models without an internet connection.
The release points to a practical shift in everyday AI use. Instead of relying only on cloud-based systems, users can try certain AI tasks on-device, including image generation, question answering, code writing, code editing, text summarizing, and text rewriting.
What Google AI Edge Gallery Does
Google AI Edge Gallery is built around a simple idea: make compatible AI models available on a phone, then let the device handle the work locally. Users can browse models, download them, and run them from the app.
The app is available for Android, and Google plans to bring it to iOS soon. For now, Android users can download Google AI Edge Gallery from GitHub by following the instructions provided there.
The app’s home screen presents shortcuts for different AI tasks and capabilities. Examples in the source include “Ask Image” and “AI Chat.” When a user selects one of these capabilities, the app shows models that are suited to that task, including Google’s Gemma 3n.
That task-first structure matters because local AI can be technical. Instead of asking users to begin with model details, the app starts with what they want to do. From there, it narrows the available choices to models that fit the selected use case.
Why Offline AI Matters
Cloud-based AI models are often more powerful than local models, but the cloud approach comes with tradeoffs. A user may not want personal or sensitive information sent to a remote data center. Another user may simply need access to AI features when Wi-Fi or cellular service is not available.
Google AI Edge Gallery addresses those concerns by running models on supported phones’ processors. Once a compatible model is downloaded, the task can run offline. That does not make every local model as capable as a cloud model, but it gives users a different option for certain workflows.
The source describes several categories of tasks the app can support:
- Generating images
- Answering questions
- Writing code
- Editing code
- Summarizing text
- Rewriting text
Those examples show the app is not limited to one narrow feature. It is more like a local testing ground for different AI capabilities, with performance depending on the device and the model being used.
Prompt Lab Gives Users More Control
Beyond the home screen shortcuts, Google AI Edge Gallery includes a feature called “Prompt Lab.” This area is for “single-turn” tasks powered by models, such as summarizing and rewriting text.
Prompt Lab includes several task templates. It also offers configurable settings that let users fine-tune model behavior. The source does not list every setting, but the feature appears aimed at giving users a more controlled way to test model responses without building a separate tool.
For developers and AI experimenters, that kind of built-in prompt space can be useful. It makes the app more than a model downloader. It gives users a place to compare how different compatible models respond to a defined task.
Performance Will Depend On Hardware And Model Size
Google warns that performance will vary. Modern devices with more powerful hardware should run models faster. Model size is another important factor.
Larger models will take more time to finish a task than smaller ones. The source gives the example of answering a question about an image. In that kind of task, a larger model may take longer to complete the response.
This means Google AI Edge Gallery is not a promise that every phone will deliver the same experience. It is an app for supported phones, and the quality of the experience depends on the balance between device power, model size, and the task being attempted.
An Experimental Alpha Built For Feedback
Google is describing Google AI Edge Gallery as an “experimental Alpha release.” That label is important. It signals that the app is early, and that Google is looking for input from the developer community.
The company is inviting developers to give feedback on the Google AI Edge Gallery experience. That feedback could matter because local AI apps depend on many moving parts: model compatibility, device performance, task design, and user expectations.
The app is available under an Apache 2.0 license. According to the source, that means it can be used in most contexts, including commercial and non-commercial settings, without restriction.
For users, the immediate takeaway is straightforward. Google AI Edge Gallery gives Android users a way to try openly available Hugging Face AI models locally, with iOS support expected later. For developers, it offers an early look at how Google is packaging on-device AI experimentation into a phone app.