Safe AI actions
The AI can speak. Your rules decide what happens next.
lunao is built to answer routine questions quickly while flagging sensitive requests for handoff or review instead of executing risky actions blindly.
Connected action execution requires explicit integrations, permissions, and production-ready approval logic.
What buyers can verify
Sources, limits, and next steps stay visible.
Fast answers are useful. Uncontrolled actions are dangerous.
Asking about a return policy is different from requesting a refund. Asking for a consultation is different from booking a confirmed appointment.
Four checks before the agent responds.
Clear public copy with current and planned boundaries.
Keep these actions out of blind automation.
Clear public copy with current and planned boundaries.
Why guardrails matter
Fast answers are useful. Uncontrolled actions are dangerous.
Asking about a return policy is different from requesting a refund. Asking for a consultation is different from booking a confirmed appointment.
Answer
Routine questions with approved source support.
Clarify
Ask a follow-up when the request is incomplete.
Hand off
Route sensitive or unresolved conversations to your team.
Draft for review
Prepare the next step where workflow surfaces exist.
Decision model
Four checks before the agent responds.
Source check
Is there approved source support?
Confidence check
Is the source strong enough?
Risk check
Is the visitor asking for a sensitive action?
Workflow check
Is there a configured workflow for the next step?
Require approval
Keep these actions out of blind automation.
Customer money
Refunds, discounts, cancellations, and payment disputes.
Orders and bookings
Order edits, address changes, booking or rescheduling.
Accounts
Private data, account updates, and identity-sensitive work.
Regulated advice
Legal, financial, medical, or regulated guidance.
Safe action docs
Use draft-and-approve boundaries before connected workflow execution.
Read guardrail docsFAQ
Common questions.
Does lunao execute actions automatically?
Not by default. Connected execution requires explicit integrations, permissions, and production-ready approval logic.
What actions are risky?
Refunds, cancellations, bookings, order edits, discounts, private account changes, and regulated advice.
Can lunao still help?
Yes. It can explain policy, collect context, and route the request to a human.
Are approval workflows fully live?
Approval-oriented surfaces exist, but production action execution should be claimed only where implemented and configured.
Build an agent that knows when to stop.
Test routine answers, sensitive requests, and human handoff before publishing your website agent.