Chion

The chart and the written summary under it come from the same frozen query in Chion, so every figure traces back to the verified SQL, not a language model.

"NASDAQ +131% vs GDP +39.6% from 2020-2025." Every number is read back from the verified result, marked SQL verified.

No summary the data doesn't support. Put it in front of a CFO without a second pass.

You can't hallucinate a query you never wrote.

Most AI analytics tools generate SQL on the fly, and sometimes it's wrong. Chion removes that failure mode by removing the generation: the verified skill you pick is the exact query that runs, frozen and never rewritten.

We reshape the output to your question. We never touch the SQL underneath.

Every question routes to one of 7 strategies, and that picks the SQL shape before any query runs.

"Top 10 customers" routes to TopK Ranked (ORDER BY … LIMIT). "Compare East vs West" routes to Comparison (parallel aggregates). "Revenue by region" routes to Dimension Breakdown (GROUP BY). Plus Find, Outliers, and All for the rest.

The shape is chosen before a line of SQL exists, so the answer is correct by construction, not guessed.

Connect Postgres read-only and 1,400+ verified queries are already sorted into Claude skills by department: Finance 312, Sales 419, Growth 248, Ops 187.

Nobody writes a prompt. Nothing touches your rows, just schema metadata and the queries your team approved.

The library exists the moment the read-only role connects.