Back to settings

AI learning

The tool watches what you correct in AI output and turns the corrections into durable style rules. Rules below are added to every prompt automatically — the more you correct, the better the next AI response.

0 pieces of feedback collected

Distill turns recent thumbs-down + correction pairs into concise style rules. Re-runs are cheap and idempotent.

Workspace-wide rules (0)

Workspace rules apply to AI output across every client. Build them by correcting outputs, then clicking distill.

How it works

  1. Every AI feature has thumbs-up / thumbs-down + a correction box.
  2. Thumbs-down with a corrected version is gold — that's a clear "here's what you should have written" signal.
  3. On distill, the AI itself analyses recent corrections and proposes durable rules ("Always use UK English", "Never start with 'In conclusion'").
  4. Rules with multiple confirmations get promoted from "tentative" → "probable" → "applied".
  5. Active rules get injected into every AI prompt for that feature (and workspace-wide rules apply everywhere).