AI for CAD Tools
Comprehensive review of free and open-source text-to-CAD tools for engineers and makers. CadQuery, OpenSCAD, Zoo.dev ML-CAD, and what each actually produces.
·
⏱
9 min read
Michelle Ben-David
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
Open-source text-to-CAD tools have reached a point where they deliver real value for makers, students, and individual engineers. CadQuery plus an LLM gives you parametric STEP output for free. OpenSCAD wrappers are great for 3D printing workflows. But for professional engineering teams, the bigger productivity lever is not generating new geometry from text. It is finding existing proven designs, accessing engineering knowledge, and avoiding duplicate work. Leo AI connects to your existing PLM stack and does exactly that.
The text-to-CAD space has a cost problem. Commercial tools charge subscription fees that make sense for enterprise teams but are prohibitive for independent engineers, hobbyists, students, and makers. If you are building a weekend project, prototyping a startup product, or teaching yourself CAD, dropping hundreds of dollars per month on generative AI design tools is not realistic.
The good news: a growing ecosystem of open-source and free text-to-CAD tools has emerged that lets you generate 3D geometry from text descriptions without spending a dime. The honest news: these tools have real limitations. Some produce parametric output but can only handle simple geometry. Others handle complex shapes but output non-editable mesh. And the AI assistance layer varies wildly from genuinely useful to barely functional.
This article reviews the open-source text-to-CAD tools actually worth your time in 2026, explains what each one produces, and helps you figure out which tool fits your specific workflow. No hype, no sales pitch, just an honest technical evaluation from an engineer who has used all of them.
CadQuery + LLM: The Best Option for Parametric Output
CadQuery is a Python-based parametric CAD library that defines geometry through code. It is not a text-to-CAD tool by itself. It is a scripting language for building 3D models programmatically. The text-to-CAD part comes when you pair it with an LLM that generates CadQuery scripts from natural language descriptions.
This combination is the strongest open-source option for engineers who need parametric, editable, production-quality output. Here is why.
CadQuery builds on the Open CASCADE geometry kernel, the same kernel underneath FreeCAD and several commercial CAD tools. The output is real BRep geometry with proper analytical surfaces, not mesh. You can export STEP files that import cleanly into SolidWorks, Creo, NX, or any other commercial CAD system. Dimensions are parametric. You can change a hole diameter by editing one variable, regenerate, and get updated geometry.
The LLM integration works like this: you describe what you want in plain English, the LLM generates a CadQuery Python script, you run the script, and you get a 3D model. Several open-source projects wrap this workflow into user-friendly interfaces. Some use GPT-4 or Claude as the language model. Others use smaller, locally-runnable models that work without an internet connection.
The output quality depends on the complexity of your request. CadQuery handles prismatic geometry beautifully: boxes, cylinders, extrusions, chamfers, fillets, patterns, boolean operations.
Where CadQuery struggles is organic shapes, complex surfacing, and multi-body assemblies. Lofts between non-parallel profiles, sweep along a 3D spline path, or complex shell operations can trip up both the language model and the CadQuery kernel itself.
For makers and engineers doing enclosures, brackets, fixtures, jigs, and simple mechanical parts, CadQuery plus an LLM is genuinely useful. It is the closest you can get to production-quality text-to-CAD output for free.
IN PRACTICE
It opens our minds to new thinking - new directions for us and for our users. We come up with better, more creative, and more efficient solutions than we did before.
"It opens our minds to new thinking - new directions for us and for our users. We come up with better, more creative, and more efficient solutions than we did before."
- Harel Oberman, CEO of Oberman Industrial Designs
OpenSCAD + AI Wrappers: The Pioneer Approach
OpenSCAD has been around for over a decade as a code-based CAD tool favored by makers, 3D printing enthusiasts, and anyone who prefers typing to clicking. Like CadQuery, it defines geometry through script rather than graphical interaction. Unlike CadQuery, OpenSCAD uses its own domain-specific language rather than Python.
Several open-source projects have wrapped OpenSCAD with AI to create text-to-CAD workflows. You describe a part, the AI generates OpenSCAD code, and you render the result. Some projects run the language model locally, making the entire pipeline free and private.
OpenSCAD's strength is Constructive Solid Geometry (CSG). It builds complex shapes by combining primitives with union, difference, and intersection operations. For parts that can be described as combinations of basic shapes with holes, cuts, and fillets, OpenSCAD works well.
The limitations are more pronounced than CadQuery's. OpenSCAD does not have proper fillet or chamfer operations. Complex curves require manual polygon definitions. And the export format is STL by default, not STEP, so the parametric nature of the script does not survive into a standard CAD exchange format.
For 3D printing projects where you will slice the STL directly, OpenSCAD plus AI is a perfectly good workflow. For engineering applications where you need STEP output or integration with a commercial CAD environment, CadQuery is the better foundation.
One genuine advantage of OpenSCAD: the language is simpler than Python, so LLMs generate correct OpenSCAD code more reliably than correct CadQuery code for basic geometry.
Zoo.dev ML-CAD (Open-Source ML Model)
Zoo.dev (formerly KittyCAD) released their machine learning-based CAD model as open source. This is a fundamentally different approach from the script-generation tools. Instead of writing code that builds geometry step by step, the ML model directly predicts 3D geometry from text input.
The results are mesh-based. You get STL-quality output, not parametric BRep. The geometry quality varies significantly with prompt complexity. Simple shapes like brackets, plates, and basic enclosures come out reasonable. Complex assemblies, thin-walled structures, and geometry with precise dimensional requirements are hit-or-miss.
The model can run locally if you have a decent GPU, which makes it fully free and private. No API calls, no subscriptions, no data leaving your machine.
Where Zoo.dev's open model excels is speed. You get geometry in seconds rather than the minutes it takes for script-based approaches to generate, debug, and render code.
Where it falls short is precision. You cannot specify exact dimensions and expect them to be honored. The model generates geometry that approximates your description, but the approximation may not be close enough for engineering use.
Practical Comparison: Which Tool for Which Job
Building enclosures, brackets, and fixtures for 3D printing: OpenSCAD plus an AI wrapper. The output goes straight to your slicer. Setup takes five minutes.
Creating parts that need to interface with commercial CAD environments: CadQuery plus an LLM. Export STEP, import into SolidWorks or Creo, and keep working. The parametric quality is high enough for engineering use on geometrically simple parts.
Quick visual mockups and concept exploration: Zoo.dev's open ML model. Fastest path from idea to visible 3D shape. Do not expect dimensional accuracy, but for rough visualization it is hard to beat.
Learning CAD through an AI-guided workflow: FreeCAD with AI assistance. You see every modeling operation in the feature tree, can edit and learn from each step, and have access to a full CAD environment for taking designs further.
Production-quality engineering with complex geometry: None of the above. If you need multi-body assemblies, complex surfacing, GD&T-ready drawings, and full PDM integration, you need a commercial CAD tool. Open-source text-to-CAD is not there yet for complex production engineering.
The Gap That Text-to-CAD Does Not Fill
Here is the honest reality for engineers evaluating these tools. Whether open-source or commercial, text-to-CAD addresses one specific step in the engineering workflow: creating initial geometry from a text description. That step might take 30 minutes to two hours in a traditional CAD tool, depending on part complexity.
But creating the geometry was probably not your biggest time sink. Finding out whether someone already designed a similar part was. Figuring out what material to use was. Checking which standards apply was. Locating the mating component's drawing to get interface dimensions was.
These surrounding tasks routinely consume more engineering time than the modeling itself. And no text-to-CAD tool, open-source or otherwise, addresses them.
Leo AI is built specifically for these tasks. It connects to your organization's full knowledge base, including PDM and PLM systems like SolidWorks PDM, Autodesk Vault, PTC Windchill, Siemens Teamcenter, and Arena PLM, and lets engineers search using natural language. Describe a part by function or geometry and find existing designs. Ask engineering questions and get answers sourced from over one million pages of standards, textbooks, and technical references.
For makers and individual engineers, the open-source text-to-CAD tools reviewed here are genuinely valuable additions to the toolbox. For engineering teams working at any kind of scale, the productivity gains from better search, knowledge access, and design reuse dwarf what text-to-CAD currently offers.
FAQ
Go Beyond Geometry Generation
Find existing designs in seconds with AI
Leo AI searches your entire vault using natural language. Find proven parts, access engineering knowledge, and stop redesigning what already exists. Free for engineers to try.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
Go Beyond Geometry Generation
Find existing designs in seconds with AI
Leo AI searches your entire vault using natural language. Find proven parts, access engineering knowledge, and stop redesigning what already exists. Free for engineers to try.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams