pgvector and operating an index

pgvector in one page

If you already run Postgres, the pgvector extension adds a vector column type and ANN indexes. The shape is below. I have not run this against a database, so check each statement against the extension's documentation for the version you install.

CREATE EXTENSION IF NOT EXISTS vector;

CREATE TABLE chunks (
  id        BIGSERIAL PRIMARY KEY,
  source    TEXT NOT NULL,
  heading   TEXT,
  body      TEXT NOT NULL,
  embedding VECTOR(1024) NOT NULL        -- must match your model's dimension
);

-- an approximate index (HNSW) using cosine distance
CREATE INDEX chunks_embedding_idx ON chunks
  USING hnsw (embedding vector_cosine_ops);

-- nearest 5 chunks to a query vector, with a metadata filter
SELECT id, source, heading, body
FROM chunks
WHERE source = 'manual-2026'
ORDER BY embedding <=> $1               -- <=> is cosine distance
LIMIT 5;

Points to remember: the vector size is fixed per column; the operator in the query must match the operator class in the index or the index is not used; and filtering plus approximate search can return fewer than k rows because the filter is applied to the candidates the index found. Test that case.

Choosing and operating an index

Always keep: the model name and version that made the vectors, the dimension, and a copy of the original text. Vectors alone cannot be turned back into text.