Building EdgeDox: Offline AI for Documents Running Directly on Your Phone

I’m a software developer interested in privacy-focused technology and running AI locally instead of relying heavily on cloud services. While experimenting with small language models and mobile inference, I noticed that most AI tools for documents require uploading files to remote servers. For sensitive documents like contracts, personal notes, or research papers, that raises privacy concerns.

That led me to start building EdgeDox.

EdgeDox is an Android app designed to run document AI directly on the device. Instead of sending documents to the cloud, the app processes them locally so files stay on the user’s phone. Users can ask questions about PDFs, summarize documents, and extract key information while staying completely offline.

The app currently runs a lightweight model (Qwen3.5-0.8B) optimized for mobile devices. One of the main challenges has been balancing performance and memory usage on smartphones, so I’ve been experimenting with smaller models, quantization, and mobile inference engines to make the experience smooth even on mid-range Android devices.

The project is still in early stages, but it has been a great experiment in pushing on-device AI forward. My goal with EdgeDox is to make document AI private, offline, and accessible, without requiring cloud processing.

If you’re curious to try it or share feedback, you can download it here:

https://play.google.com/store/apps/details?id=io.cyberfly.edgedox