I Built a Smart Bookmark Extension

The Problem

Finding Bookmarks Shouldn’t Be This Hard

Like many others, I’ve always used browser bookmarks to save interesting articles, videos, and useful websites. But over time, my bookmarks became an unorganized mess—a black hole where saved pages went to disappear. Finding a specific bookmark when I needed it felt like searching for a needle in a haystack, especially with Chrome’s default bookmark manager and most existing extensions.

The biggest issue? Search is too rigid. Sure, I could apply date filters, but that barely helped. If I didn’t remember the exact title or keyword, my bookmarks remained buried, even though I knew they were there somewhere.

So, I decided to build something that could actually solve this problem for me.

How PageStash Works

The idea was simple: use AI to make bookmarks more searchable. Instead of relying solely on keyword matching, I leveraged Gemini to create a more intuitive search experience.

Here’s how it works:

1. When searching for a bookmark, the extension gets a list of URLs and titles.
2. Gemini then analyzes the content and determines what the bookmark is actually about and filters the ones that matches your search term

For example, if you saved an article titled “10 Ways to Improve Productivity” but only remember that it was about time management, you can simply search for “tips to manage time better”, and the system will surface your bookmark.

Challenges I Faced

1. While most websites include useful metadata in the URL or title, not all do. If a webpage lacks descriptive information, the AI model might not be able to categorize it accurately, which in turn reduces accuracy.

2. Privacy Concerns
   Security was a top priority while building PageStash. All search requests are processed through Gemini’s API, meaning only Google handles your data.

To give users more control, I also introduced a “Credentials” page, allowing them to attach their own Gemini API key. This key is securely stored locally on their device, enhancing privacy while also improving response time and performance—especially when handling multiple requests at once.

Final Thoughts

While I tried to aim for a 100% accuracy, achieving perfection in AI-driven search is nearly impossible. That said, I’m proud of what I built, and I might take on the challenge of improving it in the future.

I had a blast working on this project, and I’d love to collaborate on similar ideas. If you’re interested in building something together, I’m usually free on weekends!

Source Code
Extension