Text-to-CAD, an open-source harness for AI-assisted 3D modeling, gained 73 points and 24 comments on Hacker News on May 1, 2026. The tool enables AI coding agents to generate and iterate on parametric CAD models through natural language descriptions, with full version control integration and local operation.
Parametric Source Code Rather Than Direct Model Generation
Unlike direct 3D generation tools, Text-to-CAD lets AI agents edit parametric CAD source code that lives in version control alongside project files. This approach provides full edit history with rollback capability, reproducible builds from source, AI-friendly textual representation, and professional CAD output formats. The workflow follows six steps: describe what you want to an AI agent, let the agent edit CAD source files, regenerate explicit 3D models, inspect geometry using the built-in CAD Explorer, reference geometry using stable handles for precise edits, and commit both source and artifacts to version control.
Multiple Export Formats From Single Source Files
The harness generates various file types from a single source, including STEP for industry-standard CAD interchange, STL and 3MF for 3D printing, DXF for 2D drawings, GLB for web visualization with topology data, and URDF for robot description in ROS and robotics applications. This multi-format capability makes the tool applicable to mechanical engineering, robotics, and product design workflows.
Stable Geometry References Enable Precise Iterative Edits
Text-to-CAD introduces @cad[...] syntax for stable geometry handles, allowing agents to make precise modifications like "extend the mounting hole by 5mm" without regenerating entire models. This reference system addresses a key challenge in AI-assisted design: maintaining design intent across iterations. The built-in CAD Explorer provides immediate visualization without requiring external CAD software.
Git Integration Brings Software Engineering Practices to Physical Design
CAD models live in git repositories alongside code, enabling branching and merging of design iterations, pull request workflows for design review, automated CI/CD for model generation, and historical tracking of design evolution. This integration represents a significant shift in how AI agents participate in physical design, extending the collaborative workflows familiar to software engineers into mechanical engineering and robotics domains.
Local Operation Without Cloud Dependencies
The entire system runs locally without requiring backend services or cloud APIs. All CAD generation happens on the user's machine, addressing privacy and independence concerns similar to those driving local-first coding agents. Developer Jake (earthtojake) created the project to enable AI-assisted CAD workflows with the same version control benefits that software development enjoys.
Key Takeaways
- Text-to-CAD lets AI agents edit parametric CAD source code in version control rather than generating models directly, enabling full edit history and reproducible builds
- The system exports to seven formats from single source files: STEP, STL, 3MF, DXF, GLB, and URDF for various engineering and robotics applications
- Stable geometry references using
@cad[...]syntax allow AI agents to make precise iterative edits without regenerating entire models - Git integration enables branching, pull requests, and CI/CD for 3D models, bringing software engineering practices to physical design
- The tool runs entirely locally without cloud APIs, with 73 Hacker News points indicating strong developer interest in AI-assisted CAD workflows