text2part turns a sentence into a printable 3D model. Describe one change at a time — it writes the CadQuery, runs it, and shows the result. Watch your part grow, then send it to the printer.
I sketched a part by hand, handed the photo to an AI, and it produced a nearly-correct 3D model on the first try. The trick was making that repeatable — precise, editable, mine. Many attempts later, the answer was almost embarrassingly direct: let the model write CadQuery code, one small change at a time, and show the geometry after every step.
Plain language, one feature at a time. “Give it 2 mm walls.” It refines vague prompts and asks when a request is truly ambiguous.
Editable Python you can read and tweak by hand. Primary dimensions live as named constants at the top — never buried as magic numbers.
Every step executes and renders immediately, with a timeline you can rewind. Export STL when the part looks right.
Drag a value and the whole model follows — because the code keeps its primary dimensions as named constants up top. Fix the width once, not in a dozen scattered numbers.
Generated code is untrusted, so it never runs next to your data. The app holds your key and talks to the model; a separate worker executes the geometry with no network, no secrets, and strict limits.
Nothing to set up to begin. Your first models run on our key — just describe a part. Once you're hooked, add your own key and work with no limits.
No signup, no key. Open the app, describe a part, and get the result — right away.
Sign in by e-mail and get 10 free generations on our key. No passwords — a link lands in your inbox.
Add a DeepSeek or OpenRouter key — pick provider and model in a couple of clicks. From there you only pay your own provider.