Adam launches CADAM, an open-source text-to-CAD platform
Adam (YC W25) has open-sourced CADAM, a browser-based platform that converts natural language and image prompts into parametric 3D models by generating OpenSCAD code, with exports to STL, SCAD, OBJ, GLB/GLTF, FBX, and DXF.
Score breakdown
CADAM makes parametric 3D CAD generation accessible in the browser without a desktop CAD install, and its open-source, model-agnostic architecture lets the community swap LLM backends and extend the platform toward constraint-driven modeling with build123d and CadQuery.
- 01CADAM is an open-source text-to-CAD platform built by Adam (YC W25), described as 'AI TinkerCAD'
- 02Generates parametric 3D models from natural language prompts and image references, outputting OpenSCAD code
- 03Auto-extracted parameters surface as interactive sliders; slider tweaks use deterministic regex on SCAD source — no LLM call required
Adam (YC W25) has open-sourced CADAM, a browser-based text-to-CAD platform described by founder Zach as "AI TinkerCAD." The project is built on two core beliefs: that AI will become the primary medium for mechanical design, and that generating CAD as code (text → code → CAD) is the optimal paradigm. The React app uses TanStack Start on the frontend and Supabase for auth, database, and file storage. Users can generate parametric 3D models from natural language prompts or image references; the system outputs OpenSCAD code and automatically extracts parameters as interactive sliders for instant dimension tweaking. Exports are supported in .STL, .SCAD, .OBJ, .GLB/.GLTF, .FBX, and .DXF formats.
Slider adjustments perform deterministic regex updates on the SCAD source, requiring no LLM call.
Under the hood, CADAM uses a single agentic endpoint with two modes — a parametric mode that writes and edits OpenSCAD via a `build_parametric_model` tool, and a mesh mode that generates 3D textured meshes — each swapping system prompts and tools as needed. Slider adjustments perform deterministic regex updates on the SCAD source, requiring no LLM call. The platform is model-agnostic via the Vercel AI SDK, with support for Anthropic (Claude), Google (Gemini), and OpenAI/others through OpenRouter; adaptive thinking is auto-enabled on newer models. Notably, Gemini 3.1 Pro ranked as the top model in internal evals. The entire rendering pipeline runs in-browser: OpenSCAD is compiled to WebAssembly in a Web Worker (keeping the UI non-blocking) and rendered with Three.js via React Three Fiber. BOSL, BOSL2, and MCAD libraries are supported, along with custom font support (Geist) for text in models.
Planned improvements include support for build123d and CadQuery to move beyond CSG primitives toward constraint-driven modeling, as well as better spatial context through UI for face/edge selection and viewport image integration to improve LLM spatial understanding. The repo is publicly available and open to contributions.
Key facts
- 01CADAM is an open-source text-to-CAD platform built by Adam (YC W25), described as 'AI TinkerCAD'
- 02Generates parametric 3D models from natural language prompts and image references, outputting OpenSCAD code
- 03Auto-extracted parameters surface as interactive sliders; slider tweaks use deterministic regex on SCAD source — no LLM call required
- 04Model-agnostic via the Vercel AI SDK: supports Anthropic (Claude), Google (Gemini), and OpenAI/others via OpenRouter
- 05Gemini 3.1 Pro is the top-performing model in internal evals
- 06Runs fully in-browser via OpenSCAD compiled to WebAssembly in a Web Worker, rendered with Three.js via React Three Fiber
- 07Exports to .STL, .SCAD, .OBJ, .GLB/.GLTF, .FBX, and .DXF; supports BOSL, BOSL2, and MCAD libraries
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 18, 2026 · 10:40 UTC. How this works →