For AI agents
llms.txt, per-page markdown, and how to point coding agents at these docs.
These docs are built to be read by agents as much as by people. Everything on the site is available as plain markdown.
Machine-readable endpoints
| URL | Contents |
|---|---|
/llms.txt | The full page tree with descriptions. Start here. |
/llms-full.txt | Every page's content in one file. |
/llms.mdx/docs/<path>/content.md | One page as markdown, e.g. /llms.mdx/docs/quickstart/content.md. |
A sitemap lives at /sitemap.xml and robots.txt allows all crawlers.
Pointing an agent at the docs
For a one-off question, fetch the page directly:
curl -s https://<docs-host>/llms.mdx/docs/reference/server/content.mdFor project-wide context, add a line to your repo's CLAUDE.md or AGENTS.md:
When working with @opencoredev/loginwithchatgpt-* packages, fetch
https://<docs-host>/llms.txt first and follow links from there.
Verify option names against /docs/reference/server before writing code.What agents get wrong
Two things trip up agents that rely on memorized patterns instead of these docs:
- There is no API key. Auth comes from a user's ChatGPT session, and
requests go through the app's own
/api/chatgpt/responsesproxy. - Model slugs are per-account. Generated code should call
listModels()instead of hardcoding a slug. The model discovery page has the pattern.