Hello HN! We developed a browser extension that blocks distracting sites. The killer feature is that we use LLMs and user-defined goals to provide the best possible blocking experience.

The decision-making process is not binary, we have multiple categories, allowing for more nuanced user interactions. This makes the browsing experience much less frustrating compared to other website blockers.

Sites like Reddit or YouTube are not always bad, and Wikipedia or other educational resources are not always good. By considering what the user is working on, we are able to make more informed and effective decisions.

Other blockers rely on pre-made blocklists for distracting websites. This approach is clearly insufficient for most use cases, as curating such lists is tedious—whether for the developer or the user.

We've also added quality-of-life features, including whitelists, blacklists, active hours, temporary unblocking, nuclear mode and a well-integrated (and optional) Pomodoro timer.

Learnings: Initially, we scraped page content to make decisions. However, this approach posed significant privacy concerns. Now, all our decisions are based on tab metadata, ensuring a much better level of privacy and much better performance. No page content is transferred to the server.

Data collection: We don't store browsing data unless the user explicitly reports incorrect decisions, BYOB model version will be released soon as well based on user demand.

Links: [1] https://chromewebstore.google.com/detail/hedge-ai-powered-di... [2] https://addons.mozilla.org/en-US/firefox/addon/hedge-ai-dist...