Google launches app to run AI models on phones without internet

Google has quietly launched a new experimental mobile app that allows users to download and run open-source artificial intelligence models directly on their smartphones without an internet connection.

The app, called Google AI Edge Gallery, is currently available for Android and will be released for iOS in the near future. It enables users to access a selection of models from Hugging Face, an open AI development platform, to perform a range of tasks, including generating images, answering questions, and writing or editing code. All processing is carried out locally using a device’s onboard hardware.

While cloud-based models tend to be more powerful, local execution offers advantages such as data privacy, offline access, and reduced reliance on external infrastructure. Google described the release as an 'experimental Alpha' version and is encouraging developers to share feedback. The app can be downloaded via GitHub, and is licensed under Apache 2.0, allowing for broad commercial and non-commercial use.

The home interface features shortcuts to tasks like 'AI Chat' and 'Ask Image,' each of which links to relevant models, including Google’s Gemma 3n. A built-in 'Prompt Lab' allows users to conduct single-turn tasks such as summarising or rewriting text. The lab also includes templates and adjustable settings to fine-tune model outputs.

Performance will vary based on hardware capabilities and model size, Google cautioned. Devices with more powerful processors will handle tasks more quickly, while larger models may require more time to complete operations like answering image-related queries.