Agent Surface gives automated readers a window into a Drupal site’s public content. We developed it after clients raised concerns about agent traffic making their website audiences harder to understand. Built on Markdownify, it provides Markdown versions of public pages without the surrounding website interface. Separate Markdown URLs can be cached, and repeat requests can confirm that content is unchanged without downloading it again. The aim is lighter agent access alongside the full website experience and its analytics for human visitors.
Making content easier to read also calls for clear choices about its use. Agent Surface lets site owners state separate preferences for search, AI inference and model training, with site-wide settings and content-type overrides. Those choices are published consistently through response headers, Markdown metadata and discovery files, so owners can welcome search and AI-assisted answers while declining model training. The module starts with permissive defaults and warns administrators at installation so they can review and change them.
Test-site example: Article content allows search and AI inference while declaring that it should not be used for model training.