Our summary
Ordinary page search finds the words you type, not necessarily the answer you need. Oscar Molnar built Hunch to search by meaning. Ask a question and it highlights matching paragraphs, leaving the original page visible so you can read and check the source yourself.
Hunch splits a page into paragraphs and asks Jev, an outside AI service, whether each one answers the search. It groups those questions into batches and divides long pages to fit the service's limits. Stronger matches receive stronger highlights rather than a newly written answer.
The build story is not a measured test of search quality. A high match score can still be wrong, and using the extension sends page text to an outside service. It needs an access key. The author's claims about browser and phone support were not independently tested.
Key takeaways
- Keep the original page visible so readers can check the passage instead of trusting a replacement answer.
- Split long pages and group small questions rather than assuming every page fits one request.
- Treat a strong highlight as a suggested match, not a guarantee that the paragraph answers correctly.