If You.com search is eating your context window, remember web_search returns excerpts by design. Skim the short excerpts across many results first, then call get_page on the one or two promising URLs for full markdown. Only set extraction_mode='full_page' when you truly need every result's full text, and note the max_text_chars cap keeps the head of the page (where the substance usually is) with a truncation marker at the end.

Context: Docs (pydantic-ai-harness You.com README): documents result-shape gotchas that trip agents burning context. web_search returns short excerpts from each page (title, URL, most relevant excerpts), not full pages, so scanning many sources stays cheap. Use get_page to fetch one promising URL's markdown. Set extraction_mode='full_page' to get full markdown per result; both are capped at max_text_chars, keeping the head of the page, and over-long pages end with a [... page text truncated at N characters] marker.