--- title: How Vector Search Works description: Understand how ParadeDB natively supports vector search canonical: https://www.paradedb.com/docs/concepts/vector/overview --- This is a beta feature available in versions `0.25.0` and above. Vector search finds rows whose embeddings are closest to a query vector. It is useful when relevance depends on semantic similarity rather than exact token matches. Today, many Postgres applications use the [pgvector](https://github.com/pgvector/pgvector) extension for similarity search. pgvector works well for many use cases, but separate vector indexes can struggle when a query also needs selective filters, text predicates, or frequent updates. The ParadeDB index supports pgvector's `vector` type inside the same index that stores text and columnar data. This lets ParadeDB evaluate vector similarity, full-text predicates, filters, and ranking from one index-backed query path. ParadeDB vectors use ParadeDB-built index structures designed for fast nearest-neighbor retrieval at scale. They are independent from pgvector's HNSW and IVFFlat indexes, while still using pgvector's `vector` type. Because vectors live in the same index as text and filters, vector search can be combined with [text search](/concepts/full-text/overview) into a single ranking. See [How Hybrid Search Works](/concepts/hybrid/overview). For indexing, query syntax, and tuning options, see the [Vector / Semantic Search reference](/reference/vector/overview).