A drafting desk, twice: one sketch, two desks you could actually build

A line drawing of a tilting drafting desk with a hanging pedestal, turned into two complete, interference-checked assemblies: a powder-coated steel frame with an MDF top, and the same desk in solid oak where every structural joint is a dovetail that slides together by hand. Every part is a real B-rep solid with its own STEP file; every fastener has a hole to live in.

The whole brief, verbatim

Everything below was produced from these messages and a handful of screenshots. Each one is quoted as typed; open what it became for the trimmed account of the work it triggered.

1. 1

   make the most realistic desk you can using this as a reference

   as realistic as possible for every sub assembly part so we could manufacture it

   (with a line sketch of a tilting drafting desk on cantilever legs with a two-drawer pedestal)

   what it became

   v1 designed and built: 31 part numbers in steel RHS, MDF and melamine — pinned-height cantilever legs, tilting worktop on a hinge pin with a slotted quadrant lock, welded frame, hanging pedestal with two drawers. Every part its own STEP, an AP242 assembly with product structure, BOM, viewer and a first interference check that found five real clashes and fixed them. Along the way the MCP server kept resetting on cross-hole booleans, so the work moved to the kernel's headless batch API and an assembly merger script (logged as a kernel bug).

2. 2

   let's do a separate version afterwards with real dovetail joins that would be assemblable

   what it became

   v2 generated as code (build_v2.py): the same desk in solid oak with sliding dovetails for leg↔foot, leg↔arm, rail↔arm, battens and pencil ledge, through dovetails on the pedestal carcass and drawer backs, half-blind dovetails on the drawer fronts — every male and female cut to the same 1:6 trapezoid. Notched-rack stay instead of the knob. 0.000 mm³ interference at 0/15/25°.

3. 3

   log this tessellation as an issue (figure 1), hlr side on a few surfaces is not great (figure 2)

   (with two viewer screenshots)

   what it became

   Two kernel bugs logged with the repro parts attached: the edge overlay drawing triangle diagonals across the planar faces of boolean output, and hidden-line mode drawing tessellation bands on cylindrical faces instead of one silhouette.

4. 4

   did we get fasteners in it for everything to connect it all on both versions? can we show the explosion of assembly and some sweeps for the dovetails of how that will work to assemble?

   what it became

   Honest answer was no. Every fastener was then modeled with its hole: v1 got M6 bolts and threaded inserts, confirmats, wood screws, countersunk ledge screws and hinge-pin circlips; v2 got brass butt hinges and sixteen screws. The faceted v1 tube parts were rebuilt exact with polygonal cross-holes. Exploded views for both, five animated slide-ins for the joinery, and a third kernel bug logged for polygon-prism booleans that fragment.

5. 5

   let's also get the design and BOM made up for these desks

   *review and BOM

   afterwards turn this into a showcase for the site

   what it became

   DESK_DESIGN_REVIEW.md: comparison table, what changed under review, ergonomics, structure, mechanisms, joinery map, manufacturing routes, verification, open items, and enriched BOMs (material, stock, process, mass) for both desks. Then this page.

6. 6

   get rid of the hanging support that bangs into the drawers

   (with a screenshot circling the bowed front apron)

   what it became

   Apron removed from both versions (and its housing from the oak arms); assemblies, viewers, exploded views, BOMs and the review regenerated. The rear rail carries the racking stiffness on its own.

7. 7

   does this slide through the dovetails on the one piece in this diagram?

   (with a frame from the pedestal animation)

   what it became

   It didn't — the animation slid the top onto both sides' tails at once, which the sockets don't allow. Re-animated corner by corner: top and bottom onto the left side from the right, back panel in, right side last over both boards' pins. The drawer sequence was corrected the same way.

8. 8

   got it on the site yet/

   what it became

   Page assembled in the site's shared skin (theme boot, tokens, nav strip, analytics), assets copied, hub card and nav link added across the showcase pages, pushed through deploy_site and checked live.

9. 9

   fix this text filling things up

   (with a screenshot of the viewer HUD listing every solid id)

   what it became

   Viewer HUD now says “149 bodies” instead of printing every id (logged as a kernel viewer bug).

10. 10

    put everything I prompted to get the table in the website example so they understand how much text became CAD and then this page, you can trim down your output into something that expands

    what it became

    This section.

How it was built

Each part was authored as a kernel recipe — an extrusion of a profile in a named plane, a box, a cylinder, a sweep — and cut with planar booleans: polygonal prisms for holes, tail-shaped prisms for dovetail sockets. One document per part, so every STEP holds exactly one watertight solid. The assembly is written with real product structure (one PRODUCT per part number, one NEXT_ASSEMBLY_USAGE_OCCURRENCE per instance with its placement), so a BOM reader sums quantities and a CAD system opens it as a tree rather than a soup of bodies.

The design changed under review. The first pass carried a bowed front apron under the worktop, straight from the sketch; it hung 32 mm in front of the drawer fronts and the drawers could not open, so it went. The first clash pass on v1 also found the tilt-lock stud inside the batten, the quadrant plate cutting the rear rail and the U-bolt hangers fouling the battens — each fixed and re-checked. In v2 the drawers were redesigned once it was clear that a half-blind dovetail socket must run out through the front's end grain or nothing can be slid together: an overlay front cannot be dovetailed, so v2 fronts are flush-fitted and the boxes ride on oak runners.

Exploded

How the dovetails go together

Every male tongue or tail and its female housing or socket are cut to the same 1:6 trapezoid, so the fit is line-to-line in the model (cut proud and fit by hand in the shop). The animations move the real part geometry from its approach to its seat, in the only order the joinery permits; the interference check reports each joint as face contact.

Turn it over yourself

Both assemblies in the kernel's own lightweight viewer: shaded, shaded with the real B-rep edges, hidden-line, or wireframe (keys 1–4). Left-drag orbits, wheel zooms.

Four views

Bill of materials — v1 steel / MDF

Volumes come from the kernel's mass-properties pass over the STEP solids. Masses use 7.85 g/cm³ steel, 0.75 MDF, 0.68 melamine chipboard, 0.85 HDF; the slide envelopes are counted as solid steel (real slides ≈ 1 kg for four). Not modeled: slide screws, HDF brads, the height-pin R-clip.

Bill of materials — v2 solid oak

Oak at 0.72 g/cm³, brass 8.5. Sixteen screws, two hinges, two dowels and two pins are the only hardware; everything else is joinery. Glue, drawer-bottom grooves and back-panel brads are shop details not modeled.

What the review settled

1. Ergonomics. v1 spans 695–855 mm at the front edge (seated to standing-lean); v2 is a fixed 745 mm drafting height. Tilting about the front edge keeps the pencil ledge where the hand expects it and lifts the rear by up to 275 mm.
2. Structure. Columns sit 55–60 % back on 600 mm feet, so a lean on the front edge stays inside the footprint. The v1 index plate carries the pin in double shear; v2's 1:6 sliding dovetails resist the leg being levered off the foot.
3. Pedestal. Hangs 400 mm outboard on two bars — welded steel flats in v1, dovetailed oak battens in v2 — with a rear tie to the cross rail in v1. Around 100 N·m on the bars at a loaded 25 kg pedestal.
4. Mechanisms. v1: slotted quadrant on R510.6 clamped by an M8 star knob; add a nylon washer and a serrated washer. v2: notch positions on the rack solved for 10/15/20/25° with a 200 mm strut so the dowel seats with 12 mm to spare under the top.
5. Assembly order (v2). Feet → legs → battens started in the left arm → arms → rear rail dropped in → hinges and worktop → pedestal slid onto the battens → pencil ledge slid in → racks, blocks and struts screwed. Carcass and drawers corner by corner, as animated above.

What the kernel learned

Files

More demos

All demos →

Rear differential

One wheel held, the other turns twice as fast

A complete rear differential with every bolt modelled, driven only by its mates, its feature tree open.

59 parts · 123 mates Space Program

Fly what NativeCAD builds, from an A-10 to a Saturn V

NativeCAD-built aircraft and spacecraft on a real-scale world, from a gun run to a Moon landing.

18 vehicles you can fly FoDES desk

A drafting desk built by conversation, from one line sketch

Hinge, gas struts, telescoping legs and every fastener, built by talking to an AI agent.

202 bodies, by conversation GE bracket

The GE jet engine bracket, lighter and carrying every load

The classic titanium bracket optimised and offered to print, cast, machine or weld, with STEP to download.

84.3% lighter

Build your own part with NativeCAD

Free for education and makers with a registered key · $200 per seat per year · Kernel SDK for products.

Download for Claude Desktop Get a free key Pricing Talk to sales