CADAM

CADAM is the open-source text-to-CAD web application built by Adam AI Labs, a Y Combinator-backed company. It runs entirely in the browser using OpenSCAD compiled to WebAssembly, letting users describe a 3D model in plain English and receive a fully parametric result — no installation required. The repository is licensed under GPLv3 and reached over 4,200 GitHub stars as of mid-2026.

What It Is

CADAM sits in the text-to-CAD category: it takes a natural language prompt (and optionally an image reference), generates OpenSCAD code via an AI model, and renders the result live in the browser using Three.js and React Three Fiber. The output is parametric, meaning CADAM automatically extracts adjustable dimensions and exposes them as interactive sliders. Users can export finished models as .STL, .SCAD, or .DXF files. The tool is positioned as the open-source companion to Adam's broader commercial AI workspace for hardware teams.

How the Generation Pipeline Works

The core workflow has three steps:

• Prompt input — describe the model in plain English or upload an image reference
• AI generation — the backend calls Anthropic Claude (and optionally OpenRouter or OpenAI) to produce OpenSCAD code, including BOSL, BOSL2, and MCAD library support
• Browser rendering — OpenSCAD WASM compiles the code client-side; Three.js renders the live 3D preview with orbit controls

Parameter extraction runs automatically after generation, identifying adjustable dimensions so users can tweak the model without triggering a full AI re-generation. Smart updates apply parameter changes efficiently.

Open-Source Architecture

The stack is fully documented in the repository:

• Frontend: React 19 + TypeScript + TanStack Start + Vite
• 3D rendering: Three.js + React Three Fiber
• CAD engine: OpenSCAD WebAssembly
• Backend: TanStack Start server routes + Supabase (PostgreSQL, Auth, Storage)
• AI: Anthropic Claude API (primary), with OpenRouter and OpenAI as alternatives
• Styling: Tailwind CSS + shadcn/ui

Self-hosting requires Node.js 20.19+ or 22.12+, the Supabase CLI, and API keys for at least one AI provider. Local webhook development uses ngrok for provider callbacks.

Benchmark Coverage

The repository ships a benchmarks/ directory with documented examples ranging from simple parametric parts to complex multi-part assemblies. Highlighted benchmarks include a V8 engine (22 adjustable dimensions, 8 colors), a 9-cylinder radial aircraft engine, a turbofan jet engine, a herringbone planetary gear stage, and a NACA 2412 tapered wing — each generated from a single plain-language prompt and exported as fully parametric OpenSCAD.

Update: v0.3.0

The latest release, v0.3.0, was published on June 17, 2026. The repository was created in September 2025 and has seen active development, with 551 forks and 18 open issues at the time of the last recorded update. The project's GitHub topics include text-to-cad, openscad, agents, wasm, robotics, and mechanical-engineering, signaling a direction toward agentic CAD workflows beyond simple one-shot generation.