← All models · Multilingual MiniLM
Practical · feature-extraction · WASM
Two things every global product needs: search a mixed-language knowledge base with a query in any language, and collapse "same content, different language" duplicates. Both fall straight out of one shared embedding space — no per-language index, no translation step.
One passage per line. The default set is a mixed-language FAQ. Edit it freely.
Ask in any language. The green badge marks a hit written in a different language from your query — the payoff of a shared space.
Embed every passage once, then flag pairs whose cosine similarity is above a threshold — these are likely the same content in different languages. Drag the slider to tighten or loosen.