Exterior, floors, check. You answer three screens, then read a table that puts the footprint it measured next to the area your own title block printed — before a single wall is placed.
Guided Build is the wizard inside Scope3D Desktop. It builds into the Revit session you already have open. Nothing is exported, converted, or round-tripped.
Printed area against modelled footprint, floor by floor, over a green “Everything agrees”.
Three screens, and nothing is written to the model until the third one is agreed.
Residential or commercial, and the storey height to use. Reference images — elevations, a site photo, the title sheet — are sorted out of the floor list here, so they cannot end up built as a storey.
One row per sheet: a thumbnail, an editable Revit level name, and printed area, width and depth already filled in from that sheet's own title block — labelled as read from the title block, so you check them rather than trust them. Rows reorder, and any sheet can leave the build with “This is an elevation”.
The reconciliation table: floor, printed area, width × depth, modelled footprint, and a whole-house row. Anything that disagrees blocks or warns by name. A green “Everything agrees” sits over “Start the build”.
Answered before a seven-minute build, not after.
Intake is a measurement pass, not a look at a picture. It runs before the first prompt, and it is the reason the output is something you can check rather than something you have to believe.
Vector operators
Exact wall centrelines and every printed label, taken straight off the PDF's own operator list.
Raster extraction
A scanned or photographed sheet goes through a pixel wall extractor instead.
Caps
240 segments, 400 labels. Structure stays visible instead of being buried under hatch.
Tiling
Low-DPI large sheets are also sent as overlapping tiles at 6% overlap, so a 3/32-inch dimension note stays legible.
The model converts numbers. It does not read pixels.
One program per stage, so corners resolve together. A fresh context per stage, so nothing drifts.
STAGE 01
Builds nothing. It turns the measured centrelines into sorted X and Y lists and reconciles each axis against the printed width and depth, so a disagreement costs a minute rather than a floor.
STAGE 02
Exterior walls only, as one C# program inside one transaction group. The extent is read back out of Revit afterwards and compared with what was asked for.
STAGE 03
The interior walls, off the same coordinate table the envelope was built from — so the partitions meet the exterior at numbers, not near them.
STAGE 04
Placed untagged, then counted: how many actually enclosed, plus the whole-document warning count. A room that did not close is a fact reported here, not a surprise later.
STAGE 05
Into the walls that now exist, from the openings detected and typed at intake — the same counts you read in the trace review.
STAGE 06
Last, the same way. By this point every stage has been working from the building as Revit reports it, rather than from what an earlier stage intended.
Each stage is a fresh run with an empty history and a state summary re-read out of Revit at that moment. Continuity comes from reading the building, not from remembering the conversation. Stairs are a separate opt-in stage.
Four equivalent routes in, and a whole drawing set can arrive in one go.
A file, several files, or a whole folder. Folders are walked.
An image straight from the clipboard, into the composer.
The composer's attach button, with multi-select.
Type where the file is and it reads it from there.
Every page up to twelve, so a whole drawing set arrives at once.
.txt .md .csv .tsv .json .yaml .xml .log .ifc .cs .sql, capped at 120k characters — read as text, not as pictures of text.
Rejections are named per file.
“This building has no floor slabs, no roof.” Counted in Revit, not guessed. When a run ends, a strip offers the fourteen steps in build order — and only the ones the building is actually short of.
Fix what was flagged. Floor slabs, with the stair shaft cut in the same stage. Stairs.
Posts and beams. Roof. Doors. Windows.
Elevation heights. Ceilings. Railings. Storefront.
Plumbing fixtures. Kitchen and casework. Then check it against the sheet and correct.
Each step runs on a clean context and takes a checkpoint first.
A project is one conversation about one Revit model, and it gets a real folder on your machine. Everything a run produced is a file you can open, copy, or hand to somebody else.
The project folder
%LOCALAPPDATA%\Scope3D\chats\<id>\ — one per project, created the first time you give it a drawing.
drawings\
Every sheet you gave it, at full resolution, as it came in.
checkpoints\
.rvt copies of the model, saved before a run touches anything. A guided build takes one per floor, so going back to the state before one floor is opening a file.
runs\<timestamp>-<runId>\
report.md for that run, plus every C# program it sent to Revit, numbered, as .cs files. You can read exactly what was executed against your model.
handover.md
A short note per turn. This is what makes switching model mid-project work — the next model reads the note instead of the whole conversation.
Not sure your drawing set will trace? Send it to us.
Send us a setIf yours is not here, ask us and we will answer it properly.
No, but a vector sheet is the better input. Where the PDF carries real vector geometry, the wall centrelines and every printed label come off its own operator list exactly. A scan or a photograph goes down the raster path instead, and a large sheet at low DPI is additionally sent as overlapping tiles so small dimension notes stay legible.
A scale is derived from the sheet and sanity-checked at intake, and then the Check step puts the modelled footprint next to the printed area, width and depth before anything is built. Stage 01 does the same again per axis. What it will not do is stretch geometry to make a total agree — that is explicitly forbidden, because it hides the error instead of finding it.
Yes. A guided build saves a .rvt checkpoint per floor into the project's checkpoints folder, so “go back to before the second floor” is opening one file rather than unwinding an undo stack — and it survives Revit crashing, which an undo stack does not. Then re-run that floor.
Yes, through the pixel wall extractor. You give up the exactness of a vector operator list, so the two review surfaces matter more: read the centrelines it drew on your sheet, then read the reconciliation table. Send us a scan first if you want to know before you download.
It says so, on that sheet, in the trace review — rather than producing a confident guess that looks like every other result. That is the whole point of showing the measurement before the build: a sheet it cannot read costs you seconds to spot, not seven minutes and a floor of walls.
Bring one drawing set and an open Revit session. We will walk you through a trace end to end.