July 11, 2026
Ctrl+F just got spicy
Semantic/Hybrid Search in the Browser
Blog search gets a brain, and commenters are already asking for more
TLDR: A blogger built a much lighter search tool that runs entirely in your browser and understands meaning, not just exact words. Commenters loved the speed, then instantly turned the praise into demands for multilingual support and calls for browsers to make this a built-in feature.
A humble blog search upgrade somehow turned into a mini "why can’t the web work like this all the time?" moment. The big reveal: instead of making readers download a bulky search system just to find one post, the author built a much lighter in-browser tool that can guess what you mean, not just match exact words. In plain English, that means you can search by idea, not just by typing the perfect word. For a tiny site with only 14 posts, commenters were weirdly thrilled by how fast, snappy, and frankly overqualified the whole thing feels.
The loudest reaction was pure praise. One user basically gave it the gold star treatment: "Super snappy and gives me what I’m looking for." That’s the dream for search, and in internet-comment terms, it’s practically a standing ovation. But of course the crowd immediately escalated. Instead of stopping at “nice work,” people started asking, what about multiple languages? What about giant services like Uber Eats? What if every browser just had this built in already so nobody had to mess with downloads and plumbing? That’s where the hot take energy kicked in: the comments quickly turned from “cool demo” into “why is this not a standard feature of the modern web?”
There was also some classic comment-section chaos in the form of cheerful self-promo and nerdy one-upmanship. One commenter dropped their own client-side wallpaper search project with an "xD" like they’d wandered into the talent show and decided to perform. So the vibe here is equal parts admiration, feature requests, and the usual internet move of turning one neat trick into a full-blown referendum on how broken browser features still are
Key Points
- •The article compares long-running client-side keyword search with Lunr.js against heavier semantic search systems based on sentence-transformers.
- •Bart de Goede aims to provide semantic search for a static site that runs entirely in the browser without any server or API.
- •A browser-based sentence-transformer approach using Transformers.js, ONNX, and WebAssembly was benchmarked at 23.45 MB download size, about two seconds load time, and about 18 ms per query embedding on the author's laptop.
- •The article presents model2vec 'potion' models as a lighter alternative, describing them as effectively a 4 MB lookup table.
- •The model2vec forward pass described in the article consists of tokenizing text, looking up token vectors, averaging them, and normalizing the result.