How the conversation works

Voice, menu upload, what happens when you don't know, and why it sometimes argues with you.

Talking instead of typing

Sage speaks each question aloud if you let her, and you can answer by voice, by tapping an option, or by typing. The circle at the bottom of the screen is the control: tap to talk, tap to stop. Voice is optional throughout — everything works read-and-type.

Uploading your menu

Rather than typing out every dish, photograph your menu or upload a PDF. The dishes and their stated ingredients are read off it and pre-filled. Upload before typing is the intended order.

When you do not know an answer

Say so. "I don't know" is a recorded state, not a blank — and it is never a default. The readiness check blocks finalizing on an unknown that matters for safety, so an unanswered question stops the plan being called finished rather than being quietly filled in.

Where a required fact is missing, the document prints [needs answer: …] in place of it. A plan with a plausible invention in it is worse than an unfinished one.

Why it sometimes pushes back

If your menu implies you cool food and you say you never cool anything, Sage raises it at the moment the contradiction appears rather than letting both answers sit in the same document. Being argued with by the software is the point of this part.

Knowing how far along you are

The bar across the top counts the documents your answers have built so far, and it moves when an answer unlocks one. It comes from a deterministic check of what is actually written, not from a model's estimate of how it is going.