Overview
- The experimental alpha release is available on Android via GitHub under an Apache 2.0 license with an iOS version expected
- It taps open-source models from Hugging Face along with Google’s compact Gemma3B language model for tasks like text generation
- Built on the AI Edge platform using TensorFlow Lite and MediaPipe, the app optimizes inference performance on a range of hardware
- Integrated tools such as AI Chat, Ask Image and a customizable Prompt Lab let users generate images, answer questions and write code without internet
- Performance varies by device hardware and this launch coincides with a Department of Justice probe into Google’s AI licensing agreements