Approximate vector search with vector similarity indexes

Why it matters

This feature provides efficient approximate nearest neighbor search capabilities on vector data, enabling faster similarity queries over high-dimensional vectors. It solves the problem of slow exact searches by offering a scalable, performant approach to vector similarity, thus benefiting use cases like recommendation systems, multimedia search, and machine learning model outputs.

How to use it

Users can create vector similarity indexes on their vector columns and perform approximate nearest neighbor searches using the appropriate SQL functions and index types now fully supported and stable in ClickHouse. Refer to the documentation on vector similarity indexes for syntax and usage examples.