I often find myself looking for a project that I've seen on GitHub, but I don't quite remember the name, and when I finally do find it, I notice that I've already starred the project.
Boosting starred repos may help make them easier to find, and I suspect they are also fairly often relevant to the user.
I don't know how to implement it well, however -- firstly how to integrate it well with GitHub (scrape linked accounts regularly? ehh) and secondly how to actually weigh the factor of being starred -- but I figured it's a potentially interesting idea worth at least thinking about.
The primary use case would be when searching for programming-related content where you vaguely remember seeing a useful GitHub project but can't recall its exact name. For example:
- A developer searching for "rust web framework" would see their previously-starred projects like Actix or Rocket ranked higher in results
- When looking for "react state management", starred projects like Redux or MobX would be more prominent
- Searching for "markdown parser" would prioritize starred libraries they've bookmarked for future reference
Similar features exist in other tools:
- GitHub's own search has a "Your stars" filter, but it's binary (only showing starred repos vs. showing all)
- Google allows signing in to personalize search results based on your activity
- Package managers like npm show download counts to indicate project relevance
This feature would extend Kagi's existing search ranking system by adding "starred by user" as an additional ranking factor, similar to how it currently considers factors like site quality and relevance. The boost could be moderate enough to improve discovery without overwhelming other ranking factors.
