Redis
redis-vl-dotnet Documentation redis-vl-dotnet

redis-vl-dotnet

MultiVectorQuery

MultiVectorQuery searches multiple vector fields and combines their scores into one result ordering.

Basic shape

var multiVectorResults = await index.SearchAsync(
    new MultiVectorQuery(
        [
            MultiVectorInput.FromFloat32("plot_embedding", queryVector, weight: 0.7),
            MultiVectorInput.FromFloat32("poster_embedding", [0f, 1f], weight: 0.3)
        ],
        topK: 3,
        filter: Filter.Tag("genre").Eq("crime"),
        returnFields: ["title"],
        scoreAlias: "combined_distance",
        runtimeOptions: new VectorKnnRuntimeOptions(efRuntime: 120)));

The canonical runnable reference is /examples/VectorSearchExample.

Behavior

Each MultiVectorInput requires:

  • a non-blank vector field name

  • a non-empty vector payload

  • a finite weight greater than zero

MultiVectorQuery itself requires at least one input vector and uses the same topK window validation as VectorQuery.

Score combination and projections

The library executes one vector search per input field, keeps only documents that appear in every per-vector result set, and computes the final score as the weighted sum of each per-vector distance.

Projection behavior differs slightly from VectorQuery:

  • returnFields become the projected document fields

  • the combined scoreAlias is appended automatically

  • final ordering is deterministic across equal combined scores

That deterministic merge behavior is covered in tests/RedisVL.Tests/Indexes/SearchIndexAsyncTests.cs.

Batch helpers

SearchBatchesAsync also works for multi-vector search and respects the same topK window.

Related pages