Kurvengefahr: A browser-based CAD/C

Our engineering team spent the last quarter trying to move our early-stage design reviews into a browser-based environment to stop the constant "can you screen share this" dance with the product managers. We’ve been experimenting with Kurvengefahr and a few similar browser-CAD wrappers because the friction of opening a heavy desktop client for a 5-minute tweak is where most of our momentum dies.

The goal wasn't to replace our primary PLM or high-end CAD software—that's unrealistic—but to create a "sandbox" where we can throw rough geometries and constraints without the overhead of a full project file. We integrated it into our internal dashboard so the non-engineers could actually rotate a part or check a clearance without needing a license or a workstation that sounds like a jet engine.

Adoption was a mixed bag. The junior devs loved it because it felt like using a modern web app; they started prototyping bracket placements and simple enclosures directly in the browser. The senior leads, however, spent the first two weeks complaining about the lack of precision tools they've used for twenty years. The "pushback" wasn't about the tech, but about the fear of losing "rigor." I had to frame it as a sketching tool rather than a manufacturing tool to get them to stop auditing every single line.

What actually got faster was the feedback loop. Usually, a PM asks for a change, the engineer spends 10 minutes loading the assembly, makes the change, exports a STEP file or a PDF, and uploads it to Jira. Now, we just send a URL. The "time to visual" dropped from half a day to about ten minutes.

One technical hurdle we hit was managing the constraints. Browser-based kernels can be finicky with complex dependencies. We found that if we kept the sketches simple and avoided over-constraining the geometry, the performance stayed snappy. When we tried to port over a complex assembly with 50+ parts, the browser tab started eating RAM like crazy.

For anyone trying to script basic shapes or automate some of the layout, we've been using some lightweight JS wrappers to generate the initial geometry. It looks something like this:

// Simple example of defining a basic primitive for the viewer
const part = new CAD.Box({
  width: 100,
  height: 50,
  depth: 20,
  position: [0, 0, 0]
});
scene.add(part);

The real win here is the democratization of the 3D model. When the procurement team can see the physical scale of a part in a browser tab without needing to be "onboarded" into the CAD software, the number of ordering errors drops.

The current trade-offs we're living with:

• Performance: It's great for single parts, but multi-part assemblies still lag compared to native C++ apps.
• Exporting: Moving from the browser back into the "source of truth" CAD file still requires a manual import/cleanup phase.
• Precision: You can't do high-tolerance engineering here, but for 90% of the "does this fit in the box" phase, it's plenty.

It turns out the biggest hurdle to AI and cloud-based CAD isn't the software capability, but the mental shift from "this is my private workspace" to "this is a shared live document."