AI Find Similar Lines

Keyword search needs the reference words to appear in a line. This compares meaning using sentence embeddings, so a reference like 'a young dog playing' ranks 'the puppy chased a ball' at the top even though they share no words. It scores every line by cosine similarity rather than matching strings.

No. The MiniLM embedding model runs entirely in your browser via WebAssembly. Your reference and list are processed on your device and never sent to a server. Only the model is downloaded, once, then cached.

all-MiniLM-L6-v2, a compact sentence-transformer (about 23 MB) that maps the reference and each line to a 384-dimensional vector. It is fast, widely used for semantic similarity, and runs locally through transformers.js and ONNX.

It is the cosine similarity between the reference and that line, shown as 0 to 100 percent. Higher means closer in meaning. Scores are relative, so use them to rank and compare lines rather than as an absolute pass or fail cutoff.

It comfortably handles hundreds to a few thousand lines; the reference and every line are embedded in one pass, then ranked, and the top matches are shown. Longer lists take a little longer to embed; the model downloads once on first use, then is cached, and everything runs in your browser so nothing is uploaded.