title: 'Vector & Lexical Indexes & Search' indexCards: none navigation:

• rel_notes
• installing
• configuring
• using categories:
• /edb-postgres-ai/extensions-and-tools
• /pg_extensions/ directoryDefaults: product: pgvector

This page provides an overview of vector and lexical indexing options available for Postgres in EDB‑supported distributions. Today, there are two options for vector indexes, and lexical (BM25) search support is coming soon.

Vector indexes

EDB supports two vector index options that use the vector data type introduced by the pgvector extension:

• pgvector: Provides the well‑known HNSW and IVF index types for approximate nearest‑neighbor search. It’s a widely adopted standard choice for many RAG and similarity workloads.

  • Installing pgvector
  • Configuring pgvector
  • Using pgvector
  • Official docs: pgvector README

• VectorChord: Adds its own vector index implementation, IVF‑RaBitQ, designed for large vector data volumes, lower vector search latencies, and larger vector dimensions than the standard pgvector indexes.

  • Index type: IVF‑RaBitQ (vchordq)
  • Optimized for: high‑volume datasets, low‑latency queries, high‑dimension embeddings
  • Documentation: VectorChord Index (link to be added)

Lexical search (coming soon)

EDB is adding support for lexical search via the VectorChord‑bm25 extension with a supporting pg_tokenizer extension.

• VectorChord‑bm25: Adds BM25 ranking for classic lexical search use cases.
• pg_tokenizer: Provides tokenization support for BM25.
• Documentation: BM25 install and getting started guide (link to be added)

Choosing an index

• Use pgvector (HNSW/IVF) when you want standard, widely supported ANN indexes and your vector dimensions and dataset size fit typical ranges.
• Use VectorChord (IVF‑RaBitQ) for very large vector datasets, lower query latencies, or higher‑dimension embeddings than standard pgvector indexes typically handle.
• Use BM25 (coming soon) for pure lexical search, or combine lexical and vector search for hybrid retrieval.

title: 'pgvector' indexCards: none navigation:

• rel_notes
• installing
• configuring
• using categories:
• /edb-postgres-ai/extensions-and-tools
• /pg_extensions/ directoryDefaults: product: pgvector

pgvector is an extension released as open source software under the PostgreSQL License.

pgvector provides vector-similarity search capabilities and the ability to store embeddings for retrieval-augmented generation.

For more information about pgvector, see:

• Installing pgvector
• Configuring pgvector
• Using pgvector