🛡️ Safety & control: what agents can’t do

Worth knowing explicitly:

Agents can’t exceed caps. Daily email caps, credit auto-recharge maximums, schedule timeouts - hard stops enforced by the platform, not polite suggestions to the model.

Draft-for-review means nothing sends unapproved. The approval is enforced at send time, not honor-system.

Tickets gate the judgment calls - an agent told to open tickets for spend/customers/direction will block rather than guess.

Pause always works. It’s the platform stopping the agent, not asking it to stop.

Your credentials stay yours. Secrets are write-only (readable by tools, never displayed back); AI accounts live on your infrastructure, revocable at the source any time.

💡 Tip: The trust ladder that works: draft-everything for week one → auto-handle the categories that earned it → keep tickets forever for money and customers. You choose every rung; nothing graduates itself.

And the receipts: Runs records every execution, the feed narrates everything as it happens, and every email/doc/mission an agent touches links back to the run that did it. Verify first, then trust - the product is built for that order.