Answer readers with AI chat
| Surface | What it does | Page |
|---|---|---|
| AI chat | Answers from your indexed pages and cites the page it answered from | AI chat |
| Full-text search | A keyword fallback in the header, the sidebar, or both | Search |
| Connected sources | Repositories and sites the assistant reads as fact instead of recalling them | Sources |
| Chat hooks | Your own HTTPS endpoint, run before and after the model | Chat hooks |
| Docs skills | SKILL.md files that teach any agent how to do documentation work |
Skills |
| Page feedback | Thumbs up and down on every page, so you see which answers land | Page feedback |
What the chat could not answer is recorded — unanswered questions and searches that returned nothing tell you which page to write next. Assistant answers are AI work and draw on the project balance; search, feedback, sources and hooks do not.
Where should I start?#
Connect your Sources first. A connected source is what stops the assistant answering from memory: with a repository or a site registered, "what does this cost" begins by reading it, rather than by guessing. Then turn on AI chat itself.
Related#
- Get cited by AI and found by search — the other half of the AI layer: being read and quoted OFF your own site
- Use cases — the situations teams bring to Docsbook