Skip to main content
This documentation is published in a form an assistant can read directly, so you do not have to paste screenshots or copy prose out of a rendered page. Two things are available, and they work together.

​
The index

A plain-text list of every page on this site, each with a one-line description and a link to its source. It is the file to give an assistant first: it lets one discover what exists before fetching anything.

​
The source of any page

Add .md to a page’s URL and you get its markdown source rather than the rendered page:
That works for every page here. The source keeps the headings, tables and callouts as markup, which is what makes it parse cleanly. Each one also opens with a pointer back to the index, so an assistant that lands on a single page can find the rest without being told.

​
Using them

Give the index, not the pages. An assistant that has llms.txt can fetch what it needs. One that has been pasted four pages will answer from those four and not know what it is missing. Fetch the source, not the rendered page. The rendered page carries navigation, search and styling that add noise and no meaning. Re-fetch rather than cache. These files are generated from the same pages you are reading, so they change when the documentation changes. An assistant working from a copy taken weeks ago will confidently describe a shape that has moved.