GitHub Repos intermediate 3 min read May 2, 2026
Public Preview Sign in free for the full digest →

Text to CAD: AI writes your CAD code. You get STEP files, not mesh blobs.

“Your AI coding agent can now write version-controlled CAD — 1,277 developers starred this in 10 days.”

Text to CAD: AI writes your CAD code. You get STEP files, not mesh blobs.
1 Views
0 Likes
0 Bookmarks
Source · github.com

“"You can version-control it, diff it, refactor it, and have a different agent edit it tomorrow." — per r2clickthrough.com (April 25, 2026)”

You know that feeling when you need a simple bracket, mounting plate, or custom enclosure and you either spend 90 minutes fighting a GUI-based CAD tool or wait a week for someone on the hardware team to slot it in? You already write code every day — you understand version control, diffs, and code review — but 3D geometry has always lived in a separate world of proprietary file formats with no git history. Every time you want to change a hole diameter you re-model the whole thing. text-to-cad treats mechanical geometry like code: describe what you want, get back a Python file, commit it, change it like any other source file.

cadai-agentsopen-sourcepythonbuild123droboticsparametric-design

You type a description — 'a 40mm x 20mm bracket with two M4 mounting holes 30mm apart' — inside your Claude Code or Codex session with the repo open. The AI reads the bundled skill files placed at .agents/skills/, writes Python using the build123d library, and executes it against OpenCASCADE to produce geometry. That geometry gets exported to whichever formats you request: STEP, STL, 3MF, DXF, GLB, or URDF. A local React viewer at localhost:4178 shows the result without requiring a separate CAD application. For follow-up edits, @cad[...] named handles let you reference specific geometric features so 'make the second hole 5mm larger' modifies only that feature without regenerating the whole model.

01
Parametric Python output via build123d and OpenCASCADE — you get editable Python source code, not a mesh blob, so you can diff, refactor, and re-run it without re-describing the part from scratch
02
@cad[...] geometry references — name specific features when creating a model so follow-up agents can edit 'the left mounting flange' precisely without needing the full model description repeated
03
Seven export formats from one source — STEP, STL, 3MF, DXF, GLB, URDF, and topology data all derived from a single Python file; no separate export step per format
04
Claude Code and Codex compatibility — skill files live at both .agents/skills/ and .claude/skills/ so the same repo works with either agent without reconfiguration
05
Robot Motion Skill with ROS 2 and MoveIt — if you work in robotics, you get inverse kinematics and path planning output in the same workflow as the geometry, without a separate tool
06
Local React viewer at localhost:4178 — inspect geometry immediately without installing Fusion 360 or FreeCAD; everything runs on your machine with zero cloud calls or API fees beyond your LLM subscription
Who it’s for

If you are a software developer who occasionally prototypes hardware — custom brackets, 3D-printable enclosures, or robot components — and you already use Claude Code or Codex, this removes the CAD learning curve for simple geometries at zero additional cost. Robotics engineers working in ROS 2 will find the URDF skill particularly useful for generating robot description files without hand-writing XML. This is not yet suitable for production engineering: no FEA validation, no GD&T annotations, no tolerance stack-ups, and the 2-contributor codebase carries real bus-factor risk.

Worth exploring

Worth a 30-minute setup if you are a developer who prototypes hardware and already pays for Claude Code — the MIT license, zero per-generation fees, and B-Rep output (not mesh) make the value-to-cost ratio unusually high for a 10-day-old repo. Hold off on production adoption: 2 contributors, 17 commits, no formal releases, no CI, and build123d API stability is still maturing; a breaking change in a dependency could corrupt all your stored Python CAD source files.

Developer playbook
Tech stack, code snippet, sentiment, alternatives.
PM playbook
Adoption angles, user fit, positioning.
CEO playbook
Traction signals, ROI, build vs buy.
Deep-dive insight
Full long-form analysis, no fluff.
Easy mode
Core idea, fast — when you need the gist.
Pro mode
Technical nuance, edge cases, tradeoffs.
Read the full digest
Go beyond the preview

Deep-dive insight, Easy and Pro modes, plus action playbooks — the full breakdown is one tap away.

Underrated tools. Unfiltered takes.

Read the full digest in the Snaplyze app for deep-dive insight, Easy and Pro modes, and the playbooks you can actually use.

Install Snaplyze →