Semantic search

Keyword search matches the words you typed. Semantic search matches what you meant: a query for 'cancel my plan' can find an article titled 'Ending your subscription' even though they share no words. It works by turning both the query and the documents into embeddings and finding the closest ones.

Semantic search is the retrieval step behind most RAG systems and many agent memory tools. Its weakness is exact terms: product codes, error messages, names, and rare words can be missed by embeddings alone.

For that reason many systems use hybrid search, combining keyword and semantic results, often followed by reranking. For coding agents, plain text search tools frequently beat semantic search because code is full of exact identifiers.