It reads the trip you have open
The question doesn't travel alone. The trip's id, its dates, the panel you're looking at and the id of every item on it ride along with the message, so a question about “the flight” resolves against the trip on screen rather than a memory of one. The trip's name does not ride along — it is text you typed, so Miles reads it the same way it reads everything else, off trip_read, where Stub already marks it untrusted.
It answers out of the same tools your own AI gets — trip_read, trip_budget, trip_todo_list, trip_search — so what it tells you comes off the trip's own records, not out of the model.
It reaches the trip the same way your own AI does — over MCP — but the conversation stays inside the trip you have open, not a general one you point at your data.
The assistant answering is the one holding your wallet
Ask it what a card earns on this hotel, or which one to use for the flight, and it isn't stitching an answer from two systems. Miles already holds your wallet — it's the same product — so the assistant reading the trip's dates off trip_read is the same one that knows the card.
That's the argument for asking inside the trip rather than beside it. The trip knows the flights; Miles knows the cards; the useful answer needs both, and here they are the same conversation.
Where it lives
- App
- Your trips
A language model writes the answers, so it's sometimes wrong. It answers from the trip rather than from memory, and it does not change your trip — asking it something is never a write. Check it before you act on it.
New here? Signing in creates your Miles account.