How it works
Version 0.1.0. The interface changes between updates.
From model to result
-
1 Sketch
Draw on a plane and dimension it. The sketch shows the degrees of freedom left.
-
2 Model
Extrude, cut, hole and fillet on the FreeCAD kernel, one feature at a time. Each task previews before you press OK.
-
3 Drawing
Put views and a title block on a sheet. Export it to PDF or DXF.
-
4 Study and result
Pick a study: strength, heat, natural frequencies, buckling or fatigue. gmsh meshes the part and CalculiX solves it. The result is saved with its inputs. Validation compares four textbook cases with their formulas.
What is inside
- FreeCAD, OpenCASCADE Geometry: sketches, features, assemblies, STEP, and NURBS surfaces fitted over a 3D scan.
- gmsh Meshes for the solvers.
- CalculiX Stress, natural frequencies and heat.
- MuJoCo Motion of rigs and mechanisms.
- ffmpeg Cuts and encodes video.
- Blender Renders stills and films in Cycles.
What is different
| Compared | Whet | Typical subscription CAD | Free open-source CAD |
|---|---|---|---|
| Payment | Once. Perpetual licence. | Recurring fee. Access ends when payments stop. | No fee. |
| Account | None. The key is checked on your machine. | Sign-in required. | None. |
| Files | Stay on your machine. | Varies. Some products keep them in the vendor's cloud. | Stay on your machine. |
| In one window | CAD, scan to CAD, stress, heat, motion, CAM, renders, video. | Varies by product and tier. | Varies by project. |
| AI answers | A local model. Its numbers are checked against the solver's result. | Varies by product. | Varies by project. |
On your machine
- Offline. No account. Files stay local.
- Local models run on your machine through Ollama.
- Add API models on your own key, for example through OpenRouter.
- Numbers in an answer are checked against the solver's result.
- Parts go out as STEP and STL, drawings as PDF, SVG and DXF, toolpaths as G-code.
Your work leaves your machine only when you send it: to your own server over ssh, or to an API model you added.
Running work elsewhere
Your own server
Stress and heat studies, simulations and renders can run on a machine you own, over ssh with your own key. Engine versions must match yours; the saved result names the machine.
Larger local models there
L and Ultra can run in Ollama on that machine. L needs 10.0 GB of memory, Ultra 28.0 GB.
Several models
Whet picks a model per question by task, price and past accuracy; your own rules come first. A number that fails the check is asked again from another model.
API models
Anthropic, OpenAI, xAI, DeepSeek, or any OpenAI-compatible endpoint such as OpenRouter, on your own key. Free OpenRouter models rotate when one hits its rate limit. A monthly ceiling limits spend.
Background: Story.