What is Hybrid search?

Hybrid search runs two complementary retrieval methods over the same corpus and fuses their results into one ranked list. Semantic search, using embeddings, excels at meaning: it matches paraphrases, synonyms, and conceptual questions even when no words overlap. Keyword (lexical) search, using classic algorithms like BM25, excels at exact terms: identifiers, error codes, product names, rare acronyms, and code symbols that must match precisely and that an embedding can blur together. Each method misses what the other catches, so combining them produces noticeably better recall and precision than either alone. The fusion step reconciles the two result sets, a common approach is Reciprocal Rank Fusion, which scores a document by its rank position in each list, and the merged ordering is often then passed through a reranker for a final, sharper sort. Hybrid search has become a default for production RAG and agent retrieval precisely because real queries mix conceptual intent with literal terms. Many vector databases now support it natively, storing both dense embedding vectors and sparse keyword indexes so a single query can draw on both.