Zoo Design Studio v1: A New Stack for Mechanical CAD

(Spoiler: we wrote our own geometry engine¹, programming language, ML model, and CAD GUI)

Every generation or so, someone rebuilds the geometric foundation of mechanical CAD (MCAD). It happened in 1988 with Parasolid², in 1989 with ACIS³, and, until today, the last major commercial geometry engine to reach mainstream adoption was well over a decade ago. Most modern CAD still runs on Siemens’ Parasolid from the late ’80s⁴.

After two years, fewer than 30 engineers, and a fraction of the capital poured into earlier efforts, we’re adding a new entry to that list with Zoo Design Studio v1. Below is the why, the how, and what you can expect on day one.

Why the World Needed One More Geometry Engine

Mechanical CAD’s geometric bedrock hasn’t changed much since Parasolid and ACIS defined the market. Even the most recent tools, Solidworks, Onshape and Shapr3D, still wrap Parasolid⁵⁶⁷. Industry analysts routinely talk about “only two major geometry engines” powering most MCAD software⁷. Attempts since then have stayed niche, proving how high the bar is for robustness and adoption.

But our modern roadmap demands real-time ML workflows and sub-second Boolean ops, features that don’t fit 1980-era CPU assumptions⁸. The solution: start fresh.

What Makes Our CAD App Different

GPU-Accelerated Surface/Surface Intersection (SSI)

Legacy geometry engines handle SSI with thousands of if/else branches. We derived, from first principles, this math as a parallelizable root-finding problem that runs on modern GPUs. We’ll publish detailed benchmarks as further optimizations land. That single architectural choice let a team of fewer than 30 engineers close decades of feature gap in under two years. We still have plenty of low-hanging fruit for further performance improvements and geometric features, both in the engine and in the Zoo Design Studio front-end⁹.

ML You Can Still Edit

Most AI demos, inside and outside CAD, stop at a frozen artifact you can’t tweak (like losing your layers in Photoshop). Other folks claim “Text-to-CAD” but output only a mesh. Our CAD app retains B-rep surfaces all throughout the stack: fillet an edge by clicking, adjust a dimension in code, or re-prompt with natural language, on the exact same model.

We already fine-tune our ML model for several enterprise customers. Give us your NX, Creo, CATIA, SolidWorks, or Autodesk files, anything with a feature-tree, and we’ll convert them to KCL and fine-tune a bespoke model that creates parts that look idiomatic for your organization. Sound interesting? Email sales@zoo.dev and let’s get started.

About the ML (and Why I Don’t Say “AI” Yet)

Today’s models learn from feature-tree data; they’ll misfire, and we’ll tune them. Our version of AI is the day you can type “Make me a rocket” and get a nozzle off the build plate ready for hot-fire first try. Until then, it’s ML, and it only improves with your feedback.

Code-Backed CAD (KCL)

We spent months refining the KittyCAD Language (KCL) to feel intuitive to both mechanical engineers and software developers. Dimensions use plain-English units (such as millimeters or inches, which engineers can mix and match as necessary), functions read like a feature tree, and the syntax is terse enough for human reviewable version-control diffs. But, and this is important, no mechanical engineer ever has to touch KCL. You can stay in the click-and-point UI, sketch, extrude, and assemble parts exactly as you would in any mainstream CAD tool. The code view is there only if you want deeper parametric control, automation, or to let an LLM refactor your model for you.

B-rep vs Mesh (for the Non-CNC Crowd)

Serious hardware teams stick with B-rep because sub-millimeter tolerances matter¹⁰. Meshes are great for avatars, not turbine blades and optics.

What Ships in v1

• Primitives, sketch constraints, patterns, shell, helix, extrude/revolve, fillet/chamfer, loft, sweep
• GPU accelerated constructive solid geometry operations like union, subtract and intersect
• Assemblies
• Export your models to STEP, glTF, FBX, STL, PLY, and OBJ
• Text-to-CAD ML: 40 free credits per month for every user on our Free tier
• Native app targeting macOS, Windows, and Linux (fully featured)
• Web client (currently single-file workflows)

We have a special launch price of 54% off the annual subscription to the Pro tier. With this you get unlimited Text-to-CAD credits.

Roadmap (High-Impact First)

1. Constraints 2.0 & Assembly Mates – define relationships between parts to ensure the final product always fits. We’re seeking feedback with our spec open for comments.
2. Engine-side caching – eliminate full-world rebuilds, and seamlessly restart a previous session without waiting for your part to re-render.
3. KCL caching – cache modules/geometries before even sending new data to the engine.
4. Drawings & GD&T – production-ready 2D output.
5. Conversational CAD Copilot – image-to-CAD, automatic error fixing, parameterization/refactoring.
6. Engine features – enhancements to our constructive solid geometry operations.
7. Performance – faster patterns, and other low level features, making the ML model faster at responding.
8. Code completion – using ML to autocomplete whole chunks of code in KCL.
9. And lest we forget: working through high-impact issues customers bring to us!

You can join other Zoo users and us over at the community forum where people can post their questions, concerns, ideas and suggestions!

Ready to Try It?

Today Zoo Design Studio v1 is live.

Download it, model something outrageous, and let us know where it breaks! Owning the entire stack means fixes ship fast, from geometry engine math to the button you click.

We are merely getting started on our never ending ambition. Let’s build the next forty years of mechanical CAD!