Known problems 

Please report new problems at https://github.com/web-vision/ai-llms-txt/issues. Report security issues privately as described in SECURITY.md instead.

llms.txt lists the default language only 

On multi-language sites, llms.txt always lists the page tree of the site's default language, also when it is requested under a language prefix such as /de/llms.txt. Translated pages are still available as Markdown through their own .md URLs, but are not listed.

A disabled site still answers /llms.txt 

When llmsTxtEnabled is off, /llms.txt returns a short "generation is disabled" notice with HTTP status 200, while .md requests return "page not found".

Markdown conversion 

The Markdown version is converted from the rendered HTML of the page's content, so its quality depends on the markup your templates produce:

  • Tables become Markdown tables. Markdown has no merged cells: content of cells spanning several rows or columns is kept, the layout is not.
  • Layout wrappers, <div> elements with classes such as container, row, col or grid, are unwrapped; empty elements are removed. The rest of the markup is converted as it is.
  • Images, video, audio and embedded frames are removed; links are kept.
  • Forms, interactive widgets and content rendered by JavaScript have no meaningful Markdown representation.
  • Only content elements of the page are included, from all column positions, ordered by column and then by their position in the column. The children of a b13/container grid are rendered by their container only, as on the page itself. Content that the page template renders by itself, such as menus or footer content, is not included.
  • Plugins whose output is not cacheable are rendered into the Markdown version on every request, and a page with such a plugin is not kept in TYPO3's page cache.

The result is meant to be a readable, semantic version of the content, not a faithful copy of the page's layout.