Documentation
Search and AI chat
Readers arrive with a question, not a browsing plan. Two features answer it: full-text search, and an AI chat that answers from your published pages.
Search
Full-text search across endpoints and guides, available from every page. It needs no configuration and indexes published content automatically.
AI chat
Readers can ask a question in prose and get an answer drawn from your docs.
Retrieved from your published content, not from a model's memory of an API that looks like yours.
Each answer cites the pages it came from, so a reader can verify it rather than trust it.
An uncited answer about your API is indistinguishable from a plausible guess. The citation is what lets a reader check, and what makes a wrong answer recoverable instead of a support ticket.
What it can answer
Chat retrieves from published pages, which shapes what it's good at:
| Question | Handled well? |
|---|---|
| "How do I paginate?" | Yes, if a guide covers it |
| "What does the 409 on this endpoint mean?" | Yes, if the reference documents it |
| "Why is my key returning 401?" | Only if your docs explain the failure modes |
| "What's my account's rate limit?" | No. It reads docs, not your database |
If a question isn't answered anywhere in your docs, chat cannot invent the answer, and shouldn't. Gaps in chat's answers are gaps in your documentation, which makes it a decent signal for what to write next.
Credits
Chat draws from your AI credit balance. Every plan includes an allowance, including Free.
Unpublished content is invisible
Chat and search index published pages only. A draft is not retrievable, which is the correct behaviour but occasionally surprising: a page you can see in the studio is not one chat can cite.