OpenUI
temaki.openui
Experimental tool that turns a natural-language description of an interface into live-rendered HTML, iterating on the result through follow-up instructions and converting it to framework code.
A Docker Compose file for OpenUI alone, generated from the catalog and checked by the safety inspector. Sign in and use Kitchen Credits to activate it.
Uses the Medusa cart and payment-session path. This test control does not create an order or capture payment.
OpenUI takes a description of a user interface, asks a language model to produce HTML and Tailwind markup for it, and renders the result immediately in a preview pane so the output can be judged visually rather than read as code. Follow-up instructions modify the rendered component, building up a design conversationally, and the result can be converted to React, Svelte or other framework code once it is close enough. It reaches models through hosted APIs or a local runtime, so the whole loop can run without an external provider. The project is published as an experiment rather than a supported product.
You know it worked when
- The interface loads in a browser.
- A configured model provider appears in the model selector.
- A described component renders in the preview pane.
- A follow-up instruction visibly modifies that component.
- The result converts to framework code that compiles.
Known sharp edges
- It does nothing without a configured model provider, and the container starts successfully while every generation fails on a missing key.
- Reaching a local model runtime from inside the container requires the host address set explicitly, since the default points at the container's own loopback interface.
- Generated markup is unreviewed model output and should be treated as a starting draft rather than production code, particularly around accessibility and form handling.
- The project is an experiment with irregular releases, so it is a poor foundation for anything that needs a support commitment.