Generate a robot arm CAD model from one prompt
In the past, it was almost impossible for an amateur who had never learned any CAD software to create a 3D model out of thin air, but now you only need to imagine zero […]. Task: The text-to-cad open-source project integrates with coding agents like Claude Code and Codex to generate CAD models from natural language prompts. It uses build123d to create parametric geometry, outputting standard formats (STEP, STL) and robot description files (URDF). The workflow replaces manual switching between CAD, kinematic solvers, and slicers, allowing users to generate printable parts or robot assemblies directly from text descriptions.; tools: Codex, Claude Code. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include local Python environments, build123d, and OpenCascade. The system operates offline to protect IP but requires manual correction for geometric errors and spatial inconsistencies. LLMs may hallucinate critical parameters like joint limits, posing safety risks for physical robots. Human review is mandatory before printing or deploying generated designs..
Tools
Published on Agent Work Atlas:
Overview
In the past, it was almost impossible for an amateur who had never learned any CAD software to create a 3D model out of thin air, but now you only need to imagine zero […]
Workflow
The text-to-cad open-source project integrates with coding agents like Claude Code and Codex to generate CAD models from natural language prompts. It uses build123d to create parametric geometry, outputting standard formats (STEP, STL) and robot description files (URDF). The workflow replaces manual switching between CAD, kinematic solvers, and slicers, allowing users to generate printable parts or robot assemblies directly from text descriptions.
Reported result
A 22-second demo showed a 7-DOF robotic arm design generated almost exclusively by prompts using GPT-5.5. However, community testing revealed failures, such as overlapping gussets in L-brackets and hallucinated joint limits in robot arms. The tool is currently suitable for rapid prototyping and proof-of-concept, but not for shipping engineering parts without significant human verification and iteration.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include local Python environments, build123d, and OpenCascade. The system operates offline to protect IP but requires manual correction for geometric errors and spatial inconsistencies. LLMs may hallucinate critical parameters like joint limits, posing safety risks for physical robots. Human review is mandatory before printing or deploying generated designs.
Try it
Act as a CAD engineer using the text-to-cad skill set with [replaceable: Claude Code/Codex]. I need a 3D printable bracket for [replaceable: specific component]. 1. Generate a build123d Python script defining the geometry. 2. Output the STEP and STL files. 3. Do not send to printer yet. 4. Provide a visual preview or coordinate summary for my review. 5. If dimensions are ambiguous, ask clarifying questions. 6. Once I confirm the geometry is correct, generate G-code for [replaceable: Bambu Lab printer]. Ensure all joint limits or load-bearing specs are explicitly stated in the prompt to avoid hallucination.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.