Google has quietly launched an experimental Android app called AI Edge Gallery that allows users to run artificial intelligence models directly on their smartphones without an internet connection. The app is available through GitHub and will come to iOS devices later.
Users can download and run AI models from the Hugging Face platform to perform tasks like image analysis, text generation, and coding assistance. All processing happens locally on the device, keeping personal data private and eliminating the need for cloud connectivity.
The app includes three main features: AI Chat for conversations, Ask Image for analyzing photos, and Prompt Lab for tasks like text summarization. Google’s compact Gemma 3 model serves as the core technology, processing up to 2,585 tokens per second on mobile devices.
Performance varies by device capability
Google warns that performance depends heavily on device hardware. High-end phones like the Pixel 8 Pro handle larger models smoothly, while mid-range devices may experience slower response times. The company acknowledges the app occasionally provides incorrect answers, stating it is “still under development.”
Installation requires enabling developer mode and manually installing files, making the process complex for average users. Users must also create Hugging Face accounts to download models.
The app addresses growing privacy concerns about cloud-based AI services by keeping all data processing on the device. This approach could benefit industries like healthcare and finance where data sensitivity limits cloud AI adoption. Google released the app under an open-source Apache 2.0 license, allowing commercial use without restrictions.
Sources: TechCrunch, VentureBeat