Why Mae shows her working

Every reply Mae sends is grounded in a tool call against real reservation data, and every action she takes is reversible with one tap. This isn’t a feature — it’s the contract.

Transparency by default

When Mae offers a late checkout, you can see the availability check she ran. When she logs a maintenance ticket, you can see who it routed to and undo it.

This matters for a simple reason: trust is built in small moments. Every time Mae shows her working, she is saying “I didn’t make this up — here’s where it comes from.”

What grounding means technically

Grounding means every agent response is backed by a tool call that retrieved real data. If Mae says “Room 12 is available until 2pm,” she checked. If she says “I’ve flagged your dietary requirement,” the flag is in the system.

The grounded badge in the conversation view shows which responses came from tool calls. It’s not decorative — it’s an audit trail.

Why reversibility matters

Hotels are not software companies. The person approving a refund may not be confident with technology. If an action can’t be undone, the cost of a mistake is real — a guest who was double-charged, a room that was reassigned incorrectly.

Mae’s default is: act carefully, show your work, and make it easy to correct.

The approval queue

Some actions are too consequential for Mae to take unilaterally — financial decisions, room changes, anything that affects a guest’s experience materially. These surface in the approval queue, with the evidence Mae gathered and a single-tap approve or reject.

The queue is not a failure state. It’s Mae saying “this one needs a human.”