Settings
Acme Storefront · Assisted autonomy
Autonomy level
How far the AI may carry a fix on its own before a human steps in.
Model quality
How much horsepower the AI uses. Workers run on a cheap model; oversight (plan, review, verify, judge) stays on the most advanced model — unless you change this.
Guardrails
Fine-grained permissions that hold regardless of the level above.
Let the AI apply migrations and edit rows in the connected Supabase DB.
The AI may merge a passing PR into the protected branch without a human.
Verification may run against prod (not just staging) when reproducing a fix.
If monitoring sees the same error return, reopen the task and re-run the pipeline.
After a merge, re-verify the fix against the live app via the demo login — auto-revert and reopen if it regressed.
Critical-severity tasks always pause for sign-off, whatever the autonomy level.
Spend caps
The AI pauses and asks for approval before exceeding these limits.