Type words or phrases and compute real embeddings, right in your browser, to see which ones mean similar things. A genuine model, not a fake distance. Then search them by meaning.
An embedding turns a piece of text into a vector, a long list of numbers, positioned so that things with similar meaning land near each other. “dog” and “puppy” end up close; “dog” and “car” far apart, even though they share no letters. That single idea powers semantic search, RAG, recommendations, and clustering. This tool computes real embeddings with a genuine model so the similarities you see are the real thing, not a string-matching trick.
The first run downloads a real embedding model (~25 MB, all-MiniLM-L6-v2) from a public CDN into your browser. After that it’s instant and offline. Your text is embedded locally and never uploaded, only the model files are fetched, like any web asset.
Yes. It runs all-MiniLM-L6-v2 (a real sentence-embedding model) in your browser via WebAssembly, the same family used in production semantic search. The cosine similarities are genuine.
The model weights (~25 MB) load on first use. They’re cached afterwards. Your words never leave the page, only the model is fetched.
No. Embedding happens locally in your browser. Nothing you type is sent to a server.
Read what embeddings are, where they live in a vector database, and how RAG uses them.