I built a 3D editor where you can misspell your code and it still runs

This is a fun project I made for Codex Hackathon ( and ended up winning, and using the API credits I got to make this free!! )

So I built SimuStyle, a 3D scene editor where scripts read like instructions to a friend. You can also write them badly:

when spce is presed
  make me jump up kinda high
when i bump into the coin
  give me a point

becomes

when space is pressed
  push me up 5
when I touch Coin
  change score by 1

The part I care about is what the AI doesn’t do:

• Typos are fixed on your machine. No AI call.

• AI only rewrites lines that don’t parse. Working lines are left alone and nothing extra is added.

• Every fix goes through the real compiler before you see it. If it doesn’t parse, you don’t get it.

So it’s not “AI writes your game”. It’s closer to a very forgiving compiler.

There’s also a scene assistant. Ask for “a mini-golf hole with a spinning windmill and a hole that says Hole in one!” and it builds the objects, physics and scripts, test-runs them, and lands it all as one undoable edit.

Under the hood: Svelte 5, Three.js, Rapier physics, a keyframe timeline, and Tauri 2 for the desktop build. You can export to GLB, a standalone HTML page, or WebM.

It’s free, needs no account, and runs in the browser (Windows app too).

Try it: https://simustyle.curiousengine.org/editor/
Code: https://github.com/JakeOJeff/simustyle