What is Semantic Search?

Keyword search matches strings: no shared words, no result. Semantic search matches intent, by embedding the query and the documents into the same vector space and returning the nearest neighbours. That's why it handles synonyms, paraphrase and vague phrasing that lexical search misses entirely. It has a mirror-image weakness, though — it's fuzzy where you sometimes want exact. Search for an error code or an invoice number and a keyword index wins outright. Serious systems run hybrid: lexical and semantic together, merged and re-ranked.

The obvious use is searching your documents, but the more interesting one is an agent searching its own tools. Velaris treats connectors as data in a capability mesh the agent queries at runtime: I need to move a meeting resolves to the calendar update capability without anyone having listed it in the prompt. It's the same retrieval technique pointed at a different corpus, and it's what lets the tool catalog grow past the point where stuffing every schema into context would stop working.