How it works

I built GrandpaCAD for my grandfather. He loved 3D printing but hated the software. He just wanted to make things, not fight with vertices.

You have an idea. You want to hold it in your hand. Usually, that means hours of learning CAD software or hunting for a file that almost matches what you need.

GrandpaCAD changes that. You type, it designs.

It sounds like magic, but it's really just a very smart system working hard so you don't have to. Here is what happens when you hit enter.

The process

It is a simple loop.

First, you tell us what you want. Maybe it is a "rugged case for a deck of cards" or a "replacement knob for a stove".

Then, our AI agent analyses your request, calculates the geometry, and writes the code to generate the 3D model.

Once it is done, the model appears in the viewer. You can spin it, check the dimensions, and see if it fits your needs.

Finally, you download the 3MF (or soon STL) file and send it to your 3D Printing software. Because of .3mf it also supports multi color creations, but more about that in this blog post.

Iteration

Rarely is the first draft perfect. Maybe the wall is too thin, or you forgot a mounting hole.

In traditional CAD, changing a design often means breaking it and starting the sketch over. Here, you just ask.

"Make the walls 2mm thicker." "Add a 5mm hole in the centre."

The AI understands context. It knows what it just built, so you don't have to describe the whole thing again. It just tweaks the parameters you care about.

Under the hood

That is the surface level. But for those who like to know how the sausage is made, GrandpaCAD is doing a lot more than just spitting out a model.

It searches

If you ask for a "case for an iPhone 16 Pro", the AI doesn't just guess. It searches the internet for the exact schematics and dimensions of the phone to ensure a snug fit. It grounds its creativity in real-world data.

It remembers

Design sessions can get long. You might be thirty messages deep, tweaking a complex assembly. Our system maintains a "memory" of the key details (dimensions, constraints, design choices) so it doesn't lose the plot halfway through.

It self-corrects

Writing code for 3D geometry is hard. Sometimes the AI generates code that creates invalid geometry or just doesn't compile.

We have built an automatic failover system. If a generation fails, the agent catches the error, analyses what went wrong, and tries again. It often fixes its own bugs before you even see the result.

We are also testing visual checks. The AI will soon look at the rendered model, compare it to your request, and spot issues like "this hole is covered up" or "this wall is floating" before showing it to you.

Always improving

We are building the most comprehensive evaluation framework for LLM 3D model generation. We don't just hope it works; we test it against thousands of prompts to measure success rates and geometric accuracy.

This is the first sight of what the future holds. It is good now, but it gets better every week. If you want to see how fast we are moving, check out our changelog.

Get started

You don't need to be a CAD expert to make things. You just need an idea.