BIM MONKEY
An architect takes the homeowners’ pencil sketch of a cabin to a working Revit model by handing BIM Monkey seven written briefs, one at a time. Ten recordings, the misses, and a brief template to download.
Two pages of hand sketches became a working Revit model of a one-bedroom cabin by feeding BIM Monkey seven markdown files, one at a time. No CAD tracing, no massing study, no starting from a blank template and clicking walls in by hand.
The cabin sits on a flat three-acre lot in central Washington snow country. On paper it was a pencil plan on trace, drawn by the homeowners themselves, their best attempt at a clean plan: a two-story living mass over a two-car garage, an art studio off the east side, a one-story entry wing under a shed roof, and a covered deck over the driveway. The shed roof in blue, room names scrawled in the margins. The red dimensions are mine, added so the scale was unambiguous. It’s the kind of drawing a project starts from, long before the real work begins.
The real work is usually the slow part. Getting from that sketch to a model with real wall types, real floor assemblies and a roof that actually closes takes a day or two of setup before anything interesting happens. This time I tried something different: I had Claude turn the sketch into a written build brief, split into parts, and handed each part to BIM Monkey’s Banana Chat inside Revit.
A sketch is full of decisions an AI can’t see. Which side of the wall the grid sits on. Whether the garage is heated. Where the snow goes when it slides off a 9:12 roof. Asking for “a model of this sketch” in one prompt gets you a guess. Writing those decisions down first gets you a model you can check.
So the brief reads like a set of instructions to a careful junior drafter. Every wall has a start point, an end point, a type and a reason. Every window has a host wall, a center line and a sill height. Anything scaled off the sketch or assumed carries a VERIFY tag, so I know exactly what to look at when the model comes back.
Splitting it into seven parts matters as much as the content. Each file builds on the one before it, and Banana Chat finishes and checks one part before the next one starts. If the exterior walls are wrong, I find out before 29 windows get hosted in them.
The brief also never invents a wall type. Part 01 starts by having Banana Chat list the types in our WS template and map them by Type Mark (XW6, W4, W6, IW6, F2, R1), the same marks on every WOOD studio permit set. It reports that mapping before placing anything. Stairs, railings, casework, fixtures and furniture get copied from our last project model rather than rebuilt, so the cabin inherits families we’ve already approved.
The first file sets the rules everything else follows. Four grids each way, the origin at Grid 1 and Grid A, plan north up. Grids land on the exterior face of sheathing, so the cladding can change later without moving a single dimension. Floor to floor is 11′-0", tall enough for 18" open-web floor trusses spanning 24′ over the garage. The file also ties the model to the survey: State Plane coordinates, NAVD 88 elevations, and a placeholder finish floor a foot above existing grade.
Then the exterior walls go in, listed clockwise so every wall’s exterior face points out.
Interior walls run on centerlines, except where they continue a grid line. Those align to the grid the way the exterior wall does, so the garage separation wall reads as one clean line. The garage walls get their own type with 5/8" Type X on the garage side. Rooms are placed by a point inside each one, with separation lines where the hall opens to the mudroom.
This is also where I learned to decide who moves what. The exterior walls were sitting with the grid through the middle of the stud and I wanted the outside face of the stud on the grid. I asked Banana Chat whether it should make the adjustment or whether I should, then started moving walls myself while it was still deciding. It moved them too. For a minute we were both editing the same walls, and the undo was mine. After that I told it which level was finished and which it could touch, and it did a read-only pass before moving anything.
This is where the sketch’s roof turned into geometry. A 9:12 gable over the living space, a 9:12 gable over the primary bedroom, and a 4:12 shed over the entry wing. Both gables span 24′, so both ridges land at the same height and the L-shaped roof meets in a clean valley.
Roofs are controlled by their top surface, not their bottom. When the engineer comes back with deeper rafters for the snow load, the ridge stays put and the vaulted ceiling drops a few inches. The heated slab, the floor trusses and the covered deck all get their own assemblies.
Revit crashed at the end of this part, while Banana Chat was building a family for the ice-dam components on the west roof, and I hadn’t saved as recently as I thought. I asked whether anything could be recovered. Because the whole conversation was still there, it rebuilt the floors, reattached the lower walls to the floor plate, and rebuilt the three roofs from what it had already worked out, for about two and a half dollars of API usage. My own hand edits to the ridge were gone, which was fair: it had never seen them.
The sketch had no windows, so this part is a first pass of punched openings: 29 windows and 19 doors pulled from our standard window types. L2 windows stack over L1 windows on shared centerlines. Above the shed roof, the living room gets clerestories with sills set high enough to clear the flashing. Three sliders open the art studio to the east, and two 9′ overhead doors sit side by side in the garage.
It placed all 48 openings in one pass, exactly the count in the brief. It also called the pivot door a barn door, and a few of the first-pass windows were taller than I’d draw them. That’s what a first pass is for.
The stair rises 11′ in 18 risers with open 4x12 treads. The railing is our standard steel guard with 3/8" rods and a wood cap, copied straight out of our last project model instead of rebuilt. The same railing wraps the covered deck. This part also lays out the footings, the stem walls and the posts that carry the deck roof over the driveway.
The project had no column family, so Banana Chat built a parametric post before placing the deck posts, and it sloped the garage slab without being told. The stem walls came in stepped out from the walls above, and I pulled them in by hand. The interior stair I drew myself.
The kitchen splits across two rooms, the way the sketch drew it. The range and a prep sink sit in the great room, and the main sink, dishwasher and wall ovens hide in the butler’s pantry. The rest of this part covers four bathrooms with a steam shower in the primary, a mudroom with benches and laundry, and a mechanical wall in the garage for the well tank, combi boiler and radiant manifold. Every sink, appliance and accessory is its own element, so the schedules fill themselves.
One toilet landed three feet above its floor. It took longer to notice than it did to fix.
Furniture comes in last, copied from the same project model so it’s real furniture, not boxes. It is the best check on the plan. Banana Chat places each piece, then runs clearance checks: the kitchen aisle, chair pull-out, the walk past the cars. It reports anything tight instead of moving walls. Then it sets up Walk Your Plans views for a client review.
The brief didn’t come out right the first time, and that’s the point of writing it down. Every miss showed up on paper, where it costs a sentence to fix instead of an afternoon in Revit.
The roof caught the first one. With the shed roof dying into the two-story wall, the high side of the shed landed right where the living room windows wanted to go. Dropping the entry wing’s plate to 7′-9" kept the shed low, and the living room got a row of clerestories above it. Snow was the second. A 9:12 roof sheds straight onto a 4:12 roof below it, and onto whoever is standing at the front door. The brief now calls for snow guards on the main roof, flags the shed for sliding snow loads, and keeps the entry door tucked under the low roof.
The first pass of fixtures was thin. Sinks were folded into their cabinets, the range hood was a note instead of an object, and the wall ovens were an empty tall cabinet. None of it would have shown up in a schedule. Splitting every sink, appliance and accessory into its own element fixed that. We added mirrors, shower valves and exhaust fans while we were in there.
The railings and furniture were the bigger lesson. The first draft described a steel guard in words and let the furniture default to generic families. We already had both, revised and approved, in the model from our last project. Pointing the brief at that model instead, and making Banana Chat report what it found before copying anything, took the guesswork out. Our families travel from project to project now instead of getting redrawn.
The brief also flags what it can’t solve. The combi boiler on the sketch hangs on an interior wall with no path to vent, so the brief places it as drawn and puts the question in front of me.
The model came back in about two hours and forty minutes of Banana Chat time, spread over two days, for roughly fifteen dollars of API usage. This kind of setup usually takes two to three days, call it twelve hours at $125 an hour. And it came back built on our own wall types, our own railing and our own furniture. That’s the part that matters. A fast model in someone else’s standards is just more cleanup.
If you try this, a few things make the difference. Write the decisions down before you ask for geometry. Break the brief into parts that build on each other, and check each one before you send the next. Tag every assumption so you know where to look. Decide who moves what before you both reach for the same wall. Save before you send each part. And point the brief at work you’ve already approved, so your standards carry forward instead of getting reinvented.
The sketch still matters most. Everything here started as the homeowners’ pencil on trace, and the brief only works because the intent was already on the page. BIM Monkey just closes the gap between the drawing and the model faster than I can.
Barrett Eastwood is Co-Founder and an Architect with Wood Studio in Seattle.