Schema linking, the SQL guard, how a result set chooses its own chart, and the parts of answering a database question that turned out harder than they looked.
A question deserves an answer, not a grid of rows. But asking a language model what to draw is slow, non-deterministic and unnecessary — the result set already says what it is. Here are the rules we run instead.
ReadWhatever an AI tool promises about read-only behaviour, the durable guarantee is the one your database enforces. Here is the SQL for both engines, including the grant that silently fails to cover tables created next week.
ReadThe model knows SQL better than you do. It does not know that a customer is a row in accounts, that churn lives in subscriptions.cancelled_at, or that the two join through account_id — and that is where every wrong query comes from.
ReadA tool advertising 90% on a public benchmark is telling you about that benchmark's schema. Yours has 1,700 tables and a boolean called flag_2. Here is a five-question test that predicts something.
ReadStart free, upgrade when you are asking enough questions for it to matter.
