Hacker News front pageJul 11, 2026, 08:44 UTC+8

FreeCAD runs in a browser tab, compiled from 1.5M lines of C++ to WebAssembly

FreeCAD in the Browser

magik.net ported the full FreeCAD desktop app to the browser: parametric modeling, 17+ workbenches, and an embedded Python runtime. The port took 4 days using the AI tool Fable, with all 48 prompts published. First load is ~96 MB compressed; it requires Chrome/Edge 137+ for WebAssembly JSPI, which Firefox and Safari don't ship yet. The core trick is statically linking OpenCASCADE, Coin3D, CPython 3.14, PySide6, and VTK into a single 196 MB wasm module. A hand-written 900-line fixed-function-to-WebGL2 emulator handles rendering, and JSPI suspends the wasm stack to preserve QDialog::exec() semantics for modal dialogs. The post claims this is the first time PySide6 and pivy run in a browser—I'd take that with a grain of salt since no independent verification is provided, but the technical walkthrough is detailed.

Why it matters: A 4-day AI-assisted port of a 1.5M-line C++ CAD app to the browser, with all 48 prompts published — strong engineering signal. But the CAD niche limits resonance for a general AI audience, landing right at the featured threshold.

Read the original ↗Export Markdown

Take

This one caught my eye because it hits two curiosity points at once: how you cram a massive desktop app into a browser, and how much of that grunt work an AI can actually handle. magik.net ported the full FreeCAD—parametric modeling, 17+ workbenches, embedded Python—to WebAssembly. First load is ~96 MB compressed. The core trick: statically linking OpenCASCADE, Coin3D, CPython 3.14, PySide6, and VTK into a single 196 MB wasm module, with a hand-written 900-line fixed-function-to-WebGL2 emulator for rendering, and JSPI to suspend the wasm stack so modal dialogs keep their blocking semantics.

The wilder part is the process: 4 days, almost entirely done by the AI tool Fable, with all 48 prompts and full session transcripts published. This isn't "AI wrote a CAD program"—it's "AI was used to solve a brutally specific systems integration problem": build linking, static initialization ordering, OpenGL shimming, Python extension module registration. The kind of work that normally eats weeks from a senior engineer. I'd discount the timeline a bit—only Chrome/Edge 137+ works right now, Firefox and Safari don't ship JSPI yet, and 96 MB first load isn't mobile-friendly. But if you're doing WebAssembly ports of large C++ projects or want to see where AI-assisted systems programming actually lands, the 48-prompt chain and the technical decisions it produced are worth reading closely.

Source: Hacker News front page · magik.net
Published: Jul 11, 2026, 08:44 UTC+8 · Jul 11, 2026