Guardrails let you decide exactly what your clone can and can't do on its own. Anything outside the boundaries you set is flagged and handed back to you.
What you can control
The topics and conversation types your clone will engage with
Sensitivity levels — what counts as safe to handle alone vs. escalate
Escalation rules — when a query gets routed back to you instead of answered
What your clone will always do by default
Decline financial, legal, or contractual commitments on your behalf
Avoid political, personal, or sensitive discussions
Never share confidential or personal data unless you've explicitly authorized it for that specific case
Hand off entirely for sensitive situations — see What your clone will never do
You can update your guardrails at any time as you get more comfortable with what your clone handles.