TLDR;
- Google launches AI Edge Gallery, an app that runs AI models offline on smartphones.
- Users can perform tasks like chatting, coding, and image generation without internet.
- The app taps into Hugging Face models and Google’s AI Edge tools like TensorFlow Lite.
- Currently in alpha, the app is open-source and available on GitHub for Android, with iOS support coming soon.
In a quiet but significant step forward for mobile artificial intelligence, Google has released a new app called “AI Edge Gallery,” which lets users run AI models directly on their smartphones without an internet connection.
Available for Android devices via GitHub and coming soon to iOS, the app empowers users to run a variety of AI models from the popular Hugging Face platform right from their phones.
This capability marks a shift from the typical cloud-based AI approach used by services like ChatGPT or Gemini, offering a new layer of privacy, speed, and offline utility.
Offline Functions
Once installed, AI Edge Gallery provides users with a simple interface broken into three main features: AI Chat, Ask Image, and Prompt Lab. Notably, AI Chat lets users engage in conversations with locally run language models. Ask Image allows users to upload pictures and ask the AI questions about them — identifying objects, describing scenes, or solving visual queries. On the other hand,
Prompt Lab is a sandbox environment where users can summarize text, rewrite content, or generate and debug code using “single-turn” prompts.
Tasks that were previously impossible without cloud-based AI are now accessible offline, and the app leverages powerful on-device tools like TensorFlow Lite and MediaPipe for optimized performance.
Google Boosts On-Device AI
One of the standout benefits of this local approach is data privacy. By keeping all AI processing on the device, AI Edge Gallery eliminates the need to send sensitive data to the cloud, reducing the risk of breaches or misuse.
The app is also optimized for speed, with smaller models like Google’s Gemma 3 1B , a 529MB model capable of generating over 2,500 tokens per second , running efficiently even on mid-range smartphones. While performance will vary based on device hardware, the app provides lighter models suitable for older phones as well.
Importantly, AI Edge Gallery is open-source under the Apache 2.0 license, allowing developers to experiment, adapt, and even commercialize their own builds using Google’s framework.
Big Leap Forward for Google
Currently labeled as an “experimental Alpha release,” Google is actively seeking feedback from developers and early adopters to improve the app. It can only be downloaded manually from GitHub for now, but Google says an iOS version is on the way.
This launch comes amid increased scrutiny of Google’s AI strategy, with ongoing investigations into its partnerships. Still, AI Edge Gallery represents a step toward democratizing AI, giving users full access to advanced tools without relying on cloud services.
