Search Postgres in Plain Language

Not without thinking about where the rows go. With no index and no embeddings, judging a condition means sending row contents to the Jev API, so a scan of a sensitive table puts that data off-box. Read it as a demo over a sample table rather than a drop-in replacement for an index.

The post never compares the two, so the tradeoff is unmeasured. What it does report is cost and latency for its own case: 129 rows judged in about a second for $0.0009, with a second run at 6 ms from cache. Both figures come from a single demo run on a small table.

No, and the author concedes it in the same thread, saying 129 rows is not much and that a larger run around 4,500 rows was being updated. So the headline timing describes a table small enough to fit on screen, and the bigger number was still in progress when the demo went up.

The post does not say. It shows the SQL surface, a plain WHERE jev(table, 'condition') call, and claims no index or embeddings are needed, but install steps, supported Postgres versions, and whether the planner can still optimize around the predicate are all left to the pg-jev repository README.

Zachi, who says he does cool things at ActianCorp and lists a security background of thirty-eight CVEs in his profile, and who maintains pg-jev on GitHub. That context is worth noting because the demo is a database-tooling experiment from someone with database and security experience rather than a research release.

One function inside a WHERE clause: WHERE jev(people, 'could work from home') or WHERE jev(people, 'name sounds european'). No index or embeddings are involved, so the extension classifies each row against the plain-language condition. The examples are subjective conditions, which is the part to think about before pointing it at real tables.

There is a live demo at pgjev.zachi.dev, linked from the thread and described as querying Postgres in plain English. It is separate from the GitHub repository, so you can see the behaviour before setting up the extension, though the post does not say what data the live demo runs against.