Multi-Vector

Overview

Use multiple embeddings per document for enhanced retrieval.

Store Multiple Embeddings

-- Create table with multiple embeddings
CREATE TABLE documents (
    id SERIAL PRIMARY KEY,
    content TEXT,
    title_embedding vector(384),
    content_embedding vector(384),
    summary_embedding vector(384)
);

Multi-Vector Search

Search across multiple embeddings:

-- Search with multiple vectors
SELECT id, content,
       multi_vector_search(
           embed_text('query'),
           ARRAY[
               title_embedding,
               content_embedding,
               summary_embedding
           ],
           ARRAY[0.2, 0.6, 0.2]  -- weights per embedding
       ) AS combined_score
FROM documents
ORDER BY combined_score DESC
LIMIT 10;

Learn More

For detailed documentation on multi-vector strategies, embedding selection, weight optimization, and performance tuning, visit: Multi-Vector Documentation

Related Topics

• Hybrid Search - Combine multiple search types
• Vector Search - Vector similarity