Multi-Vector Search (Muvera)

Weaviate Studio supports Weaviate's named-vector / Muvera search — query across multiple vector spaces simultaneously and combine results with configurable strategies.

Version Requirement

Multi-vector search requires Weaviate v1.26+ for near queries and v1.27+ for hybrid queries. A version-gate notice is shown on older servers.

Target Vector Selection

The Vector Options drawer shows all named vectors for the current collection:

• Each named vector displays its vectorizer badge (e.g., text2vec-weaviate, none)
• Auto-selection — all named vectors are pre-checked when opening a multi-vector collection
• Deselect individual vectors to narrow your search

Join Strategies

When multiple target vectors are selected, choose how distances from different vector spaces are combined:

Strategy	Description	Best For
Minimum (default)	Conservative merge using minimum distance	General-purpose, safe default
Sum	Add distances from all spaces	When all vector spaces matter equally
Average	Mean of all distances	Balanced multi-modal search
Manual Weights	Per-vector distance weighting with sliders	Fine-grained control
Relative Score	Weight by relative scores	Score-based combination

Weight Editor

Available when using Manual Weights or Relative Score strategies:

• Per-vector sliders for fine-grained distance weighting
• Normalize button to auto-balance weights
• Real-time weight sum display

Search Modes

Multi-vector search works with all vector search modes:

• Text (Semantic) — requires a text vectorizer on the collection
• Similar Object — find objects similar across multiple vector spaces
• Raw Vector — provide vectors for each target manually

Copy as Code

Generated TypeScript and Python code snippets automatically include the correct multiTargetVector combination call:

// TypeScript example
collection.query.nearText('your query', {
  targetVectors: ['title_vector', 'body_vector'],
  combination: 'minimum',
});

# Python example
collection.query.near_text(
    query="your query",
    target_vectors=["title_vector", "body_vector"],
    combination="minimum"
)

Validation Guard

The Run Vector Search button is automatically disabled when:

• A multi-vector collection has no target vectors checked
• A clear validation hint explains the issue

This prevents cryptic server errors and guides users to the right configuration.