Multilingual Vectorization

Libro doesn't just understand English. It natively speaks over 50 languages using our paraphrase-multilingual-MiniLM-L12-v2 core model.

How it works

The vector space maps semantics, not just vocabulary. This means the concept of "Software Engineer" in English is mapped incredibly closely to "Ingénieur Logiciel" in French.

Example

If you ingest context in Japanese:

await ctx.ingest({
  userId: "user_jp",
  content: "私は犬が好きです" // "I like dogs"
});

And later query it in English:

const result = await ctx.getContext("user_jp", "What is their favorite animal?");

Libro will perfectly resolve the semantic match and return the Japanese text. No pre-translation required.