Beginner1 min readUpdated August 1, 2026
Why can't you just use Postgres for vector search?
Stub — heading outline only, full article to follow.
Yes, pgvector exists — what it actually gives you
Where exact and IVF-based search in Postgres starts to strain
What purpose-built vector databases optimize for that Postgres doesn't
Operational tradeoffs: one system vs. two
A reasonable rule of thumb for when to switch
Related questions
What exactly is an embedding, and why does AI engineering need them?
Embeddings are numeric vectors that represent the meaning of data — text, images, audio — so that similar things end up close together in vector space.
How does ANN search trade accuracy for speed, and where's the knob?
ANN search trades a small amount of recall for a large speedup, and every major index type exposes a specific parameter that controls exactly how much.
What breaks first when you go from a million vectors to a billion?
Memory is usually the first wall, followed by index build time and tail latency — each requires a different fix, and they don't all show up at once.