PrivyDocs

Inspiration

We realized that most professionals — especially lawyers, doctors, and privacy-conscious users — hesitate to use online AI tools because their documents often contain confidential information. As a result, they lose out on the productivity benefits that AI provides. We wanted to bridge that gap by making AI assistance possible entirely offline and on the user’s own device.

What it does

PrivyDocs is an offline AI-powered document assistant that runs locally inside Google Chrome. It uses Chrome’s built-in Gemini Nano models to summarize, proofread, rewrite, translate, and redact text — all without sending any data to external servers.
Documents are stored securely in IndexedDB, and the app is available as a Progressive Web App (PWA) for seamless offline access.

How we built it

We built PrivyDocs as a lightweight web application using Vanilla JavaScript and Web Components, with IndexedDB for local storage and a Service Worker for full offline support.
The app communicates directly with Chrome’s built-in AI APIs — including the Prompt, Summarizer, Writer, Rewriter, Proofreader, and Translator APIs — which are powered by the Gemini Nano on-device model.
We also added a Redaction Engine that detects personally identifiable information using regular expressions and allows one-click redaction, plus an Export System to generate PDF or DOCX versions locally.

Challenges we ran into

Access to experimental Chrome AI APIs: Many APIs are still in early release or Canary-only, so configuring flags and model readiness was a challenge.
Offline AI integration: Ensuring the AI worked without network calls required deep understanding of how Chrome loads on-device models.
Data privacy and performance: We carefully optimized for instant local response while keeping everything secure in the browser sandbox.

Accomplishments that we're proud of

Achieved a completely offline AI workflow — no cloud, no external APIs.
Built a fully functional local-first document workspace with auto-save, AI actions, redaction, and export.
Created a clean, responsive UI using native Web Components — no heavy frameworks.

What we learned

We learned how powerful the Chrome Built-in AI stack can be when combined with local web technologies, and how privacy-preserving design can coexist with AI productivity.
We also gained valuable experience in debugging and validating on-device model behavior through chrome://on-device-internals.