pmndrs creator releases ClassCAD: AI-driven embeddable CAD — describe a part in plain language, the agent writes a program with verifiable geometry
Paul Henschel (@0xca0a, author of zustand/react-spring/r3f/pmndrs) announced the release of ClassCAD, an AI-driven embeddable CAD his small Swiss team polished for years. Four-step flow: describe (plain language) → execute (the agent writes a ClassCAD program with variables/loops/expressions against api.v1, api.tree(), api.graphic(), run live via one run_script) → verify (deterministic renders + calculateMassProperties producing volume/COG/extents as numeric proof, sections/views/first- and third-angle drawings) → embed (headless C++/WebAssembly engine in browser/server/desktop/mobile, wrappers for JS/Node/Python/.NET, live parameter updates in @buerli.io/react). Ships an MCP server (@classcad/mcp for Claude Code/Codex/Cursor/VS Code/OpenCode) and an in-app assistant (@buerli.io/ai with any tool-calling LLM); the knowledge layer documents 264 API methods, each verified against a live engine, with BM25 retrieval. All packages open on GitHub (awv-informatik/classcad-ai).
