pgvector

pgvector is a Postgres extension that adds a vector column type and nearest-neighbour operators, so embeddings live next to the relational data they describe instead of in a separate system.

pip install -qU langchain-postgres

from langchain_postgres import PGVector
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings(model="text-embedding-3-small")

vector_store = PGVector(
    embeddings=embeddings,
    collection_name="my_docs",
    connection="postgresql+psycopg://user:password@localhost:5432/mydb",
)

vector_store.add_documents(documents)

Why put vectors in Postgres

If your application already stores its relational data in Postgres, pgvector avoids running and operating a second database, keeps backups and transactions unified, and lets you join a vector search against ordinary SQL filters instead of maintaining metadata filtering logic in two places.

Index types

Index	Build time	Query speed	Recall	Use it when
None (exact)	none	slow at scale	exact	small dataset, correctness matters most
IVFFlat	fast	good	approximate, tunable	moderate dataset size, simpler tuning
HNSW	slower	fastest	approximate, high recall	larger dataset, query latency matters

tip

Start with no index (exact search) while the collection is small — an approximate index is an optimization for scale, not a default you need on day one.

See also

• Overview — the shared VectorStore interface.
• Comparison — pgvector against the alternatives.