AI Chat

Docsbook ships an AI chatbot trained on the content of your specific documentation. Readers ask a question, the bot searches relevant pages, reads them, and returns a streamed answer with citations.

How it works#

The chat uses a RAG flow built on tool calls. The model decides when to invoke each tool, and the UI shows the live trace of every step.

Search — the model queries the indexed graph of your docs for relevant sections.
Reading — it opens specific pages and quotes the text it needs.
Answer — it composes the response and streams it to the reader via streamdown.

Each call is logged so you can later inspect what was asked, what was retrieved, and whether the answer was useful.

Customization#

You control how the bot behaves without writing backend code:

Suggested questions — seed the empty state with 3–5 starter prompts.
System prompt — replace the default instruction set with your own voice and rules.
Pre/post-LLM hooks (PRO) — transform the prompt before the LLM sees it and post-process the response before the reader sees it. See Chat Hooks.

Providers#

The default provider is OpenRouter with openai/gpt-4o-mini. You can switch to any of:

providers:
  - openrouter
  - openai
  - gemini
  - anthropic

Bring your own API key from the workspace settings. Costs charged by the upstream provider are billed to your key, not to Docsbook.

Limits#

Plan	AI requests / month
Free	Chat disabled
PRO	200
PRO+	2000

Workspaces on PRO that exceed the monthly cap can continue at $0.01 per request.

Why it matters#

Teams using Docsbook AI Chat deflect 847 support tickets per month on average — readers find answers without opening Slack or email.

Pricing#

AI Chat requires the PRO plan ($150 lifetime) for 200 requests/month, or PRO+ ($59/month) for 2000 requests/month.

Chat Hooks — Inject and post-process the LLM pipeline.
MCP Server — Manage chat settings from Claude Code.