The Reviewer

AI that catches AI's mistakes.

A second, cheaper model reads what the first AI produced, checks it against your checklist, and blocks it when something is wrong — showing exactly what it caught. The verdict is per item; the block is decided in code, not by vibes. Works with any model, and learns from the mistakes you correct.

Try one:

What to check

Source ground truth — optional

AI output what to review

Checklist one rule per line

Reviewer which model does the checking

Model optional

API key Anthropic · used for this request only, never stored

Use a small, cheap model — the point is that it still catches the big one's mistakes.

Verdict

Load an example or paste your own, then run the review. The reviewer grades each checklist item and blocks the output if any item fails. Disagree with a verdict? Teach it — your correction is applied to every review after.

What it has learned

0 lessons

Nothing yet. When you mark a verdict wrong and save the rule, it shows up here and is sent with every future review so the reviewer applies it. Stored in your browser only.

Want this guarding your AI in production?

Book a 30-min call →