Try Gemma on your phone with AI Edge Gallery¶
AI Edge Gallery is Google AI Edge's app for running open models directly on a phone or laptop. Every inference happens on the device — no network, no account, nothing leaves the hardware — which makes it the fastest way to see what Gemma 4 actually feels like on the device you carry.
It is the no-code counterpart to LiteRT-LM: the same runtime underneath, wrapped in an interface you can hand to someone who does not write Python.
Install¶
| Platform | Requirement |
|---|---|
| Google Play | Android 12 and up |
| App Store | iOS 17 and up |
| macOS | Download from the releases page |
Without Google Play access, install the APK from the latest release.
What you can do with it¶
- AI Chat with thinking mode — multi-turn conversation, with a toggle that shows the model's step-by-step reasoning. Thinking mode works with the Gemma 4 family.
- Ask Image — point the camera or pick a photo and ask about it.
- Audio Scribe — transcribe and translate voice recordings in real time.
- Prompt Lab — a workspace for single-turn prompts with direct control over temperature and top-k, useful for tuning before you commit settings to code.
- Agent Skills — augment the model with tools such as Wikipedia grounding, maps and visual summary cards. Skills can be loaded from a URL.
- Mobile Actions and Tiny Garden — offline device control and a natural-language garden game, both powered by a fine-tune of FunctionGemma 270M.
- Model management and benchmarking — download models, side-load your own, and benchmark each one on your actual hardware rather than trusting someone else's numbers.
Bring your own weights
Gallery loads custom models, so a Gemma you fine-tuned and exported with LiteRT-LM can be tested on-device without building an app around it.
How it fits together¶
| Layer | What it does |
|---|---|
| Google AI Edge | Core on-device ML APIs and tooling |
| LiteRT | The runtime that executes the model |
| Hugging Face | Where the .litertlm weights come from |
Next steps¶
- Run the same models from a terminal or your own app with LiteRT-LM.
- Fine-tune first, then export: fine-tuning overview.
- Read the project wiki for detailed setup, including corporate-managed devices.