Vector Search

MySQL Vector Search and AI Integration

MySQL 9.0 introduces the VECTOR(dimensions) data type, enabling semantic search and RAG architectures without a separate vector database.

VECTOR Type (MySQL 9.0+)

CREATE TABLE docs (
    id        INT AUTO_INCREMENT PRIMARY KEY,
    content   TEXT,
    embedding VECTOR(1536) NOT NULL,
    VECTOR INDEX idx_emb (embedding)
);

-- ANN similarity search
SELECT id, content,
       1 - VECTOR_DISTANCE(embedding, STRING_TO_VECTOR(?), 'COSINE') AS similarity
FROM docs
ORDER BY VECTOR_DISTANCE(embedding, STRING_TO_VECTOR(?), 'COSINE') ASC
LIMIT 10;

RAG Pattern

1. Embed the user's question
2. Vector search MySQL for top-K relevant chunks (similarity > 0.7)
3. Build prompt: context + question
4. Send to LLM and stream response

Alternatives

• MySQL 9.0 VECTOR: zero ops overhead, good for <5M vectors
• pgvector: HNSW index, better performance, requires PostgreSQL
• Milvus/Qdrant: billion-scale, dedicated ops cost