micheleriva/zbsearch

ZBSearch is a JavaScript search engine, a piece of software you can add to a website or app so users can search through data, whether that data lives in a browser, a server, or elsewhere JavaScript runs. The README explains it as a fork of an earlier project called Orama, restarted by the same team of engineers after they left the original company, now maintained independently without outside investors. It supports full text search, meaning matching plain words and phrases, as well as vector search, which compares numeric representations of meaning called embeddings rather than exact words, and a hybrid mode combining both. Other listed capabilities include filtering results, searching by geographic location, pinning specific results to the top for merchandising purposes, grouping results into facets, boosting certain fields to matter more in ranking, tolerating typos, and a ranking method called BM25. It also supports stemming and tokenization, ways of normalizing words for matching, across 30 languages, plus a plugin system for extending it. Installing it is done with common JavaScript package managers like npm, yarn, pnpm, or bun, or by importing it directly in a browser using a script tag pointed at a CDN. Using it starts by creating a database instance and defining a schema describing what fields your data has, such as strings, numbers, booleans, geographic points, and vectors. You then insert documents matching that schema and run searches against them. For vector or hybrid search you need to supply numeric embeddings representing the meaning of your data at search time. The README mentions a companion plugin, plugin-embeddings, for generating those embeddings automatically if you do not already have a way to produce them, using TensorFlow.js as the underlying engine.