Z2lkOi8vSm9iYmVyL1F1b3RlLzYzODczNDA2

On site, the 📱 Walkthrough screen is the fast way — big buttons, rooms, photos, no reloads. This box does the same parse from the desk: "Room 1, master bedroom, wall 1 is 10x12, walls 2 and 3 are 12x12, each wall has trim and the ceiling needs to be painted…" It also understands named walls and room footprints ("the kitchen is 12 by 14 with 9 foot ceilings"). Your API key is set, so natural rambling parses well; the built-in parser is the fallback.

Length × width × height fills the takeoffs: walls = 2(L+W)×H, ceiling/floor = L×W, trim = perimeter. Override any number if the room isn't a box.

On site, snap photos straight from the 📱 Walkthrough screen — they tag themselves with whichever room is active. Here, every photo has a room dropdown — photos are grouped by room below, and untagged ones are outlined so nothing stays ambiguous. The AI builder trusts these tags when attributing work to rooms.

One collective analysis of everything on this quote — the photos, the dictation transcript, the rooms with their measurements, and the lines you've already entered — proposing the complete line list from the price catalog. Nothing is added until you tick it.

Sends the attached photos, your price catalog and the current lines to Claude, which answers as an estimator: what it sees in each photo, which catalog items apply, whether the quote already covers it — and what looks missed. It only ever suggests; you decide what gets added.