Inspiration

My friends and I noticed Microsoft injecting AI overlays on YouTube to summarize videos. Since most people prefer Chrome over Edge, we wanted to create a privacy-focused alternative that works entirely offline that it's respecting user privacy while providing powerful AI summarization for web content.

What it does

Gemini Presto is a Chrome extension that summarizes text articles and blog posts using Google's Gemini Nano AI model that all processing happens offline on your device. The extension features:

  • 4 summary types: Key Points, TL;DR, Teaser, and Headline formats
  • 3 length options: Short, Medium, and Long summaries to fit your needs
  • Multiple export formats: Plain text, Markdown, and Rich Text
  • Summary history: Save, search, and manage all your summaries
  • Enhanced citations: Automatically captures author, date, and source context
  • Complete privacy: No data leaves your device - everything runs locally

The extension integrates seamlessly with Chrome's side panel, providing a clean interface that doesn't interfere with your browsing experience.

How we built it

We built Gemini Presto using modern web technologies and Chrome's latest AI capabilities:

Technologies:

  • Chrome Extension Manifest V3 with Side Panel API
  • Google Chrome's built-in Summarizer API
  • Gemini Nano model running entirely on-device

Architecture:

  • Content scripts for text extraction from web pages
  • Service worker for background processing
  • Local storage for summary history and settings

Challenges we ran into

  • Content Extraction: Different websites structure their content differently. We had to develop robust text extraction algorithms that could identify main article content while filtering out navigation.
  • Multimedia Limitations: Video and audio summarization wasn't possible due to Chrome's permission limitations and DRM restrictions.

Accomplishments that we're proud of

We created an AI extension that can read articles and blog posts without ever sending user data to external servers. Everything happens on-device. It works.

What we learned

Chrome AI works locally on our hardware. Their API documentation is simple and shows how to use it with a few lines of code.

What's next for Gemini Presto

If Google enables Chrome to load custom models, we plan to expand Gemini Presto to support multi-modal content including videos, images, and document formats while enabling custom model integration for specialized use cases.

Built With

Share this project:

Updates

Submission history