Search

Similarity, filters, and reranking in one call.

Search sends a query vector and an optional filter. Add a rerank stage when the first pass is quantized or hybrid.

hits = db.search(
    "notes",
    vector=query,
    filter={"tenant": "acme"},
    rerank="late-interaction",
    limit=8,
)

Multivector points accept a map of named vectors. Choose which names vote on each query.