Google has rolled out a new app: AI Edge Gallery. Launched on May 31, this experimental app puts AI on your smartphone with no cloud, no internet, and no data sharing with Big Tech.
The app is live on GitHub for Android, with an iOS version expected soon. It runs entirely offline using Google’s Gemma 3n models for tasks ranging from image analysis to code writing.
The app targets developers and features three main tools: AI Chat for conversation, Ask Image for visual tasks, and Prompt Lab for quick tasks like rewriting text. Users can download models like Gemma-3n-E2B and Qwen2.5-1.5 B from platforms like Hugging Face. 
Some users on Reddit questioned its novelty, likening it to existing options like PocketPal. Others expressed security concerns. Nevertheless, the app’s hosting on Google’s official GitHub alleviates impersonation fears, with no malware reported yet.
Testing on a Samsung Galaxy S24 Ultra showed promising results. The largest Gemma model is about 4.4 GB, while the smallest is 554 MB. Once downloaded, it operates offline without needing further data. Performance is reminiscent of early GPT-3.5, delivering decent responses without a fast connection.
For basic tasks, the app performs well. It has a context window of 4096 tokens, feels limited by 2025 standards but simulates older chat experiences effectively. Privacy is its standout feature—ideal for healthcare or journalism where sensitive data is involved.
Battery drain concerns loom, especially with larger models. Setup complexity might deter casual users. Limited model support and lack of .safetensor integration are drawbacks. However, Google‘s shift mirrors a growing demand for local AI processing.
This app provides an essential resource for privacy-conscious users, if the issues surrounding model access can be addressed.
Google’s AI Edge Gallery marks a significant step. If they add .safetensor support, it could become vital for AI enthusiasts looking to safeguard their data.
