Postgres for AI Apps: A Practical Primer

Most AI apps don't need a dedicated vector database. Postgres with the pgvector extension handles relational data and similarity search in one place.

Quick answer: install pgvector, add a vector column, index it with HNSW or IVFFlat, and query with the <-> distance operator — no separate vector database to operate.

Why one database beats two

Running a vector database alongside Postgres means two systems to back up, secure, and keep in sync. pgvector collapses that into one.

CREATE EXTENSION IF NOT EXISTS vector;

CREATE TABLE documents (
  id serial PRIMARY KEY,
  content text,
  embedding vector(1536)
);

CREATE INDEX ON documents USING hnsw (embedding vector_cosine_ops);

Where it starts to strain

At very large scale (tens of millions of vectors, sub-10ms latency requirements), a dedicated vector database may still win — but that's a later problem for most projects, not a starting one.

Key Takeaways

• Start with Postgres + pgvector; it's almost always enough at first.
• HNSW indexes trade a bit of build time for much faster queries.
• Fewer moving parts means fewer things that can silently drift out of sync.

Related Reading

• What Is RAG, and Why It Matters