Niches · CAD, engineering & fabrication
Model a new part from a description with dimensions
No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.
What counts as done
A STEP file (plus STL or 3MF for printing) whose bounding box and hole count, diameters and positions match the stated dimensions, with the source model left editable.
Measured, for our own tests: dimensions within 0.1 mm, within 5 minutes and under 10 calls. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: make a bracket 50 mm wide with four M5 holes
, design an enclosure for this board
, model a part that fits these measurements
Does the job, with a limit 4
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Adam C Adam AI Labs, Inc. | A task delegated to Adam's agent models in a connected CAD application and files come back through the task, with STEP, STL and 3MF previewed. No dimension check, time bound or call count is documented, the surface is beta, and acc-20 found directional prompts ambiguous. source |
| Autodesk Platform Services E Autodesk, Inc. | No description-to-model route: the agent must write an add-in or script in Inventor's or Fusion's desktop API and run it on the Automation API's cloud engine. The Design Automation guide renders empty for readers, so time bounds and call counts could not be read. source |
| Onshape C PTC Inc. | An agent can add modelling operations or write FeatureScript into a new Part Studio and export STEP with units, keeping the model editable; acc-28 moved to this route. No description-to-model service exists, so dimensions rest on the agent, and calls count against 2,500 a year on Free and Standard. source |
| Zoo C Zoo Corporation | Text-to-CAD and the Zookeeper agent return editable KCL, conversion gives STEP and STL, and bounding-box and volume endpoints check the result. acc-02's bracket prompt ran 8 minutes then failed, acc-03's cached outputs were unreachable (both still open), and acc-05 was told assemblies fail. source |
Doesn't do this job 3
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| nTop E nTopology Inc. | ntopcl runs an existing notebook with JSON inputs and returns JSON outputs; it does not build a new part from a description, and there is no hosted geometry API. source |
| PrusaLink D Prusa Research a.s. | A printer interface; it takes sliced G-code and makes no models. source |
| SimScale E SimScale GmbH | SimScale imports geometry for simulation; geometry primitives exist only as references inside a simulation spec, and no part is modelled or exported. source |
What stops agents finishing
- job runs for minutes then fails with an internal error
- assemblies and fasteners refused or wrong
- model returns geometry that looks right but misses stated dimensions
- output only as mesh
Around it
Before: chat.user-request, vision.read-sketch. After: cad.queue-print, cad.quote-or-order-part, cad.simulate-part.
Where this comes from
Weight 30 of 100 in CAD, engineering & fabrication: 14 of 37 accounts (Zoo 6, Adam 1, code-CAD with CadQuery, build123d and FreeCAD 6, one eval); the most discussed job on Hacker News
The same as data: /api/v1/use-cases/cad.part-from-description.json, or the MCP tool get_use_case with id cad.part-from-description. Other jobs in this category: /api/v1/niches?category=cad.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.