LLM-Friendly Output experimental

VitePress can generate LLM-friendly versions of your documentation at build time. To enable it, add the following to your .vitepress/config.js:

export default {
  llms: true
}

This emits the following files into the output directory alongside the regular HTML build:

• llms.txt — an index of all pages with links and descriptions, for LLMs to discover your documentation.
• llms-full.txt — the entire documentation in a single markdown file.
• A raw markdown version of each page (e.g. /guide/getting-started.md next to /guide/getting-started.html), with <!-- @include --> directives expanded and route rewrites applied.

The table of contents in llms.txt follows your sidebar structure and order when a sidebar is configured.

Options

Pass an object to customize the output:

export default {
  llms: {
    // toggle generation while keeping the rest of the options,
    // e.g. only generate on CI (defaults to true)
    enabled: !!process.env.CI,

    // used to build absolute links; falls back to sitemap.hostname.
    // links are root-relative when absent
    hostname: 'https://example.com',

    // defaults to the index page's hero name, then the site title
    title: 'My Project',

    // defaults to the index page's hero text, then the site description
    description: 'Documentation for My Project',

    // glob patterns for pages to exclude from LLM output, matched
    // against source paths (relative to srcDir) and output paths
    ignoreFiles: ['about/team.md', 'changelog/**']
  }
}

Targeting Humans or LLMs

Wrap markdown in <llm-only> to make it visible only in the generated markdown output, or in <llm-exclude> to keep it out of it:

<llm-only>

Extra context that only appears in the markdown served to LLMs.

</llm-only>

<llm-exclude>

Interactive demo that only makes sense in the browser.

</llm-exclude>

Content in <llm-only> is removed from the rendered HTML pages, and content in <llm-exclude> is removed from the generated markdown. The tags themselves never appear in either output. They are only processed when the llms option is enabled.

Each tag must be alone on its own line, like in the example above. Tags inside fenced code blocks or anywhere else on a line (e.g. in inline code) are left untouched, so you can document them.

Limitations

• Only the root locale is emitted — translated locales are skipped.
• Pages generated from dynamic routes are skipped.
• <<< code snippet imports and image references are left as-is in the markdown output.