Skip to main content
Hybrid search combines two or more retrieval signals so your results benefit from each. Full-text search (BM25) and sparse-vector search match specific terms like product codes, error strings, and names. Semantic search (dense vectors) matches on meaning, so a query still finds an answer phrased with different words. Each method misses what the others catch, and hybrid search closes that gap.

Combine signals

The following patterns differ in where the signals meet. A filter can serve as one of them: metadata filtering narrows the candidate pool before ranking, so an out-of-scope document can’t compete for a result slot. RRF is a fusion method, not a synonym for hybrid search. It’s one way to merge ranked lists, while “hybrid search” is the broader approach of combining signals. See Reciprocal rank fusion.

Set up an index

For a new document or text workload, use the Documents API. Declare one or more string fields with full_text_search, then add the dense side:
  • Integrated embedding: Add embed to a string field, so Pinecone indexes the text for full-text search and embeds it for semantic search. See Integrated embedding.
  • Your own vectors: Declare a dense_vector field alongside the full-text fields. See Bring your own vectors and the full-text search example schemas.
If you converted a legacy vector index that held both dense and sparse vectors, the new index has a separate field for each, plus any string field you gave full_text_search. The legacy combined dense and sparse query doesn’t carry over, because a document search ranks by one scoring type per request. Combine them with one of the patterns above. For a legacy vector index, see Hybrid search on a single index and Hybrid search on separate indexes.