UI generation with a working feedback loop.

Standard syntax reduces UI adaptation

You do not need to learn KUI’s features first. Describe the interface, style, data and actions with HTML/CSS web standards. An agent can consult current SDK docs, choose Lua or native host integration and check the result in a working template.

Real runtime feedback after generation

Build errors, DOM, computed styles, box models, input results and screenshots provide inspection material. An agent can inspect the running KUI app through MCP; developers can review the same code, documentation and acceptance results.

Check whether the first UI is usable

A useful first UI needs successful startup, correct visuals and key interactions. Preserve the first submitted artifact and record later changes separately. Templates and inspection guidance are available here; rates across a fixed task set, with model and context details, require measurement.

• Agent guide and task template →
• Get the native workspace project →