Turn Mintlify and Fumadocs pages into plain Markdown

Pass Agno's new DocumentationMarkdown to Knowledge.sync_pages as its transform, and it rewrites each page before Agno chunks and embeds it. Documentation sites serve their pages as MDX full of <Steps>, <Tabs> and <Warning> tags. Without the transform, those tags end up in your chunks and embeddings, and your agent quotes them back to users.

from agno.knowledge.page import DocumentationMarkdown
 
report = await knowledge.async_sync_pages(
    url="https://better-auth.com/docs/llms.txt",
    transform=DocumentationMarkdown(profile="fumadocs"),
    index_version="docs-v1",
)

DocumentationMarkdown converts steps, tabs, callouts, cards, fields, media and wrapper components into Markdown. It unwraps components it doesn't know, keeps code fences and indentation exactly as written, and drops the "Documentation Index" preamble that these sites put at the top of every page.

Pick the profile that matches your docs site:

DocumentationMarkdown never runs JavaScript. It treats attribute expressions as literal strings and leaves inline JSX alone. You can map your own component names onto the built-in ones with component_aliases, or render them yourself with component_renderers. To convert a single string outside a sync, call normalize_mdx, which agno.knowledge.page also exports.

Changing the transform changes what you store, so bump index_version when you add it to an existing index.

See the cookbook, and learn more about synchronizing published Markdown in the documentation. For a production example, see how we built the Agno Docs Agent.