Docs
Use oks-ui with AI
The whole library, in a form an assistant can read in one request.
What’s published
/llms.txt— a curated index: the key facts, and links to every section./llms-full.txt— the entire library as one Markdown file: install instructions, every component with its prop table, the design tokens, the FAQ. Regenerated on every deploy, so it never drifts from the real API.
Both are plain text/plain, linked from the site's <head> and referenced in robots.txt, so crawlers and agents find them on their own.
The workflow
- Open a chat with ChatGPT, Claude, or any assistant that can fetch a URL.
- Paste the prompt below, filling in the screen you want.
- The assistant reads
llms-full.txt, learns the API, and writes real oks-ui code — correct component names, correct props, token-based styling. - Paste the result into your project. It compiles.
Starter prompt
You're helping me build a UI with oks-ui.
First, read https://www.oks-ui.com/llms-full.txt — it has every component,
every prop, and the full design-token reference in plain Markdown.
Then build: <describe the screen you want>.
Rules:
- Use only oks-ui components.
- Style with --oks-* CSS variables, never hard-coded colours.
- Import each component from its subpath, e.g. import { Button } from "oks-ui/button".Why it works
LLMs are good at pattern-matching an API once they can see it, and bad at guessing one they can't. A single authoritative file — generated from the same data that drives this site — removes the guessing. Every page here also ships clean semantic HTML and per-page structured data, so pasting any individual component or pattern URL works too.