Google has made its new open-weight Gemma 4 models available to run entirely on smartphones through the Google AI Edge Gallery app, allowing users to chat with an AI assistant, ask questions about photos and use voice input without an internet connection.
The app, available for Android and iOS, downloads a compact version of Gemma 4 to the device and runs it locally using the phone’s own processor. Google says no prompts, images or audio leave the phone, which addresses one of the main privacy objections to cloud-based assistants. There is no subscription: the models are released under an open license and the app is free.
What the app does
According to Google’s developer documentation, the on-device Gemma 4 variants accept text, images and audio as input. Users can hold a spoken conversation, point the camera at an object or document and ask about it, or run text tasks such as summarizing and drafting. Google also highlights the models’ ability to follow multi-step instructions and call functions, which it describes as agentic skills, meaning the model can be used as the brain of a small automated assistant on the device itself.
Because everything runs locally, response speed depends on the phone’s hardware. Android Authority’s coverage notes that the smaller Gemma 4 variants are intended for recent mid-range and flagship devices, and that performance on older phones will be slower. Google positions AI Edge Gallery partly as a demonstration for developers, who can use the same model files and runtime in their own apps.
Why on-device matters
Running models on the phone rather than in a data center has three practical advantages: it works without connectivity, it avoids per-query cloud costs for both the user and the developer, and it keeps personal data on the device. The trade-off is capability. A model small enough to fit on a phone cannot match the largest cloud models on difficult reasoning tasks, so on-device assistants are best suited to everyday work such as summarization, translation, quick answers and image understanding.
Gemma 4 is the latest generation of Google’s open model family, which the company offers alongside its proprietary Gemini models. By pairing the release with a ready-to-use mobile app, Google is making local AI accessible to ordinary users rather than only to developers willing to set up their own tooling, and signaling that it expects more AI processing to shift from the cloud to the devices people already own.
