Qdrant: search

Search documents from a given Qdrant vector database collection. Supports plain dense search and hybrid retrieval: in Hybrid Search mode a sparse prefetch and a dense prefetch are fused with Reciprocal Rank Fusion (RRF). The sparse vector is computed on the fly from the request's query string using the configured sparse method (BM25 or SPLADE); Search dense only bypasses the sparse leg.

Pre-requisite: Install Qdrant application Profile > {Organization} > Applications to grant Zparse access.

Parameters

Applicationrequired

Select configured Qdrant application.

Collectionrequired

Which collection to use.

Collection moderequired

Match the mode used when indexing this collection:

• Single Embedding — plain dense search.
• Hybrid Search — dense + sparse prefetches fused with Reciprocal Rank Fusion (RRF), for better keyword recall.

Dense Vector Name

Hybrid Search only — name of the dense vector when the collection uses named vectors.

Sparse Vector Name

Hybrid Search only — name of the sparse vector when the collection uses named vectors.

Sparse Method

Hybrid Search only — algorithm used to derive the sparse query vector from the query text:

• BM25 — classic lexical scoring (default).
• SPLADE — learned sparse representation.

Search dense only

Hybrid Search only — when enabled, skip the sparse prefetch and run a named-dense search over Dense Vector Name. Useful for purely semantic queries when the collection still uses named vectors.

Input

Search queryrequired

Embeddings search request with model like:

{
   "query": <String>, # Actual text to generate embeddings from.
   "query_vector": [f32, f32, ...], # Search vector
   "filters": [ # Filter for metadata
     {
         "attribute": <String>, # which attribute to filter from
         "value": <String>, # which value matching given attribute to filter from
     }
   ],
   "limit": <Number>, # Own many vector to return from rag database
   "exact": <Bool>, # Whether to exact match on filters of not
}

Output

JSONrequired

Embeddings request with model like: Embeddings search request with model like:

{
   "data": [
     {
       "score": <Number>, # which rating score given document got from original search vector (higher is better)
       "identifier": <String>, # document identifier
       "payload": <JSON>, # document content
     }
   ]
}