essay
Why agent memory belongs in Postgres
A vector store is an index, not a database. Agent memory needs both — and pgvector is the door.
The default move when someone says “agent memory” is to reach for a vector database. It feels right — embeddings in, similar things out, done. It’s almost never right.
A vector store is an index, not a database. It can find similar things; it can’t transact, join, version, or enforce row-level security on what it returned. Agent memory is none of those problems in isolation — it’s all of them at once. You need recall, ground truth, ergonomics, and the operational maturity of something a bank already trusts.
That’s Postgres. With pgvector for similarity, pg_trgm for fuzzy text, RLS for multi-tenancy, and 30 years of operational discipline behind it. Bento is the small, opinionated kit of tables and an HTTP API that turns those primitives into something an agent can actually use.
The full deep-dive lands later this month. For now: if your agent has goldfish memory, you don’t need a new database. You need to use the one you already have.