Wayfare

You type your meetings, or paste them as prose and confirm the draft. A constraint planner enumerates every itinerary that satisfies the hard constraints, then AI layers rank them and assess risk. Booking re-searches every leg against live flight data before anything is ticketed, and out-of-policy trips route to a manager.

An empty result is useless. If no itinerary is feasible, the planner returns the constraint that blocked it and the smallest change that would fix it.

Money and time. Prices are stored as integer cents and every time is timezone-aware, because a flight that crosses midnight in another timezone is exactly where bugs hide.

Each AI layer has a deterministic fallback, so the planner still works when the model is down. The model makes it better, it isn't holding it up.

Flight search and booking run against Duffel's test environment. A live token is refused outright, so it can never buy a real ticket by accident.

Policy answers are grounded in the company's policy document, not the model's memory.