pgvector

pgvector

pgvector is an open-source PostgreSQL extension for vector similarity search. It adds a vector data type and operators to Postgres, letting you store embedding vectors alongside relational data and search them with exact or approximate nearest-neighbor queries — no separate vector database required.

Key Features

• Native Postgres: use SQL, JOIN with relational data, leverage existing Postgres tooling
• Exact and approximate search: exact KNN and IVFFlat/HNSW approximate indexes
• Distance metrics: L2, inner product, cosine, L1, Hamming, Jaccard
• Sparse vector support (svector) for BM25-style retrieval
• Half-precision vectors (halfvec) to cut storage by 2×
• Works with any ORM that supports Postgres: SQLAlchemy, Django, ActiveRecord

Quick Start

CREATE EXTENSION vector;

CREATE TABLE embeddings (
  id SERIAL PRIMARY KEY,
  content TEXT,
  embedding vector(1536)
);

CREATE INDEX ON embeddings USING hnsw (embedding vector_cosine_ops);

-- Nearest neighbor search
SELECT content, 1 - (embedding <=> '[0.1,0.2,...]') AS similarity
FROM embeddings
ORDER BY embedding <=> '[0.1,0.2,...]'
LIMIT 5;

from pgvector.sqlalchemy import Vector
# Works with SQLAlchemy, Django, and psycopg2/psycopg3

Install via ai-supply

npx ai-supply add pgvector-postgres-embeddings

Curated mirror of the open-source pgvector (PostgreSQL License). Get it from the source.