# Documentation endpoints for AI (/en/docs/guide/guides/llms-txt)

The llms.txt, llms-full.txt, per-page Markdown and Accept-header negotiation this site provides.



*For: developers · operations*

This page is about **making this documentation readable by an AI** — not about using the library.
To wire x-iztro into an AI application, see
[Letting an AI read the chart](/en/docs/guide/guides/llm).

## `/llms.txt` [#llmstxt]

A structural index of the site, listing the title, description and link of every page in that
language. Good for letting a model locate a page before fetching it.

<Endpoint path="/en/llms.txt" />

## `/llms-full.txt` [#llms-fulltxt]

The full Markdown text of the whole documentation set in that language — one fetch gives a model
complete context.

<Endpoint path="/en/llms-full.txt" />

## Both endpoints are per-language [#both-endpoints-are-per-language]

Index and full text exist once per language and are never mixed. The full text is meant to be
dropped into a context window wholesale, and mixing in a language you cannot use only crowds the
window.

| Endpoint                                  | Contents                                                                                                   |
| ----------------------------------------- | ---------------------------------------------------------------------------------------------------------- |
| `/en/llms.txt` · `/zh/llms.txt`           | The structural index for that language, listing the other languages and the full-text endpoints at the end |
| `/en/llms-full.txt` · `/zh/llms-full.txt` | The full text for that language                                                                            |
| `/llms.txt` · `/llms-full.txt`            | The root paths that the llms.txt convention specifies; contents identical to the `/zh/` pair               |

## Per-page Markdown [#per-page-markdown]

Append `.md` to any documentation page URL to get that page's Markdown source:

<Endpoint path="/en/docs/guide/guides/config.md" />

Both `.md` and `.mdx` work and return the same content. The "Copy Markdown" button under a page
title fetches exactly this endpoint, and the "Open" dropdown next to it can send the page straight
into ChatGPT or Claude.

## It also works without knowing the suffix convention [#it-also-works-without-knowing-the-suffix-convention]

When an AI agent requests any documentation page, declaring a preference for Markdown in the
`Accept` header returns the Markdown source instead of the full HTML page:

```bash
curl -H "Accept: text/markdown" <site>/en/docs/rust/palace
```

For that one page, the HTML is around 400 KB and the Markdown around 14 KB.

<Callout>
  All three routes return plain text with no navigation, styling or scripts — cheaper in tokens than
  having a model scrape HTML, and more accurate.
</Callout>

## Suggested usage [#suggested-usage]

When asking an AI about x-iztro, hand it `/en/llms-full.txt` as context. The full text for a single
language is around 500 KB, comfortably inside the context window of any common model.

If you only care about one topic, fetching the relevant single-page `.md` is cheaper:

| Topic                       | Page                                          |
| --------------------------- | --------------------------------------------- |
| Parameters and installation | `/en/docs/guide/getting-started.md`           |
| Domain concepts             | `/en/docs/guide/concepts.md` and its subpages |
| Boundaries and schools      | `/en/docs/guide/guides/config.md`             |
| Field dictionary            | `/en/docs/guide/data-model.md`                |
| to\_text format             | `/en/docs/guide/guides/to-text.md`            |