We are building an assistant that answers questions about coding agents the way the manual does: grounded in the corpus, citing its sources, honest about freshness. It is not live yet, and we would rather ship nothing than a chat box that hallucinates fixes.
Search-grounded answers. Every response is assembled from problems, recipes, and configs in the manual — with citations linking straight to the pages, so you can check the evidence and its verification counts yourself.
Version awareness. Tell it your agent version and it will prefer fixes verified on that version, and warn you when the best answer it has is marked outdated.
“I don't know” as a feature. When the manual has no verified answer, it will say so and point you at filing a problem report instead of improvising one.
Press ⌘K anywhere to search topics, agents, frameworks, and tools. It is fast, and the results are the same pages the assistant will cite.
A hand-ordered reading path through the fundamentals — the pages we would want an assistant to summarize for a newcomer anyway.
Open the collection