AI Assistant
AI Assistant gives users accurate, contextual answers drawn from your entire knowledge base — not just your docs
AI Assistant gives your users fast, accurate answers about your documentation using natural language. It's personalized to your users, can be embedded into your website or product, and can appear in the sidebar or search box of your published docs.
Think of it as a product expert available to all of your users, in the places and times they need it most.
The Assistant uses agentic retrieval to understand the context of queries based on the user's current page, previously-read pages, and previous conversations.
Try asking the Assistant a question in the box below:
Configure AI Assistant
To configure AI Assistant, open Customize, under Tools in the site sidebar, and click AI Assistant. Here you can enable the Assistant and customize its instructions, welcome message, and suggested questions.
Add custom instructions
Custom instructions adjust the Assistant's tone of voice and add product-specific knowledge or terminology, so its answers consistently match your wider style.
Custom instructions refine the Assistant's tone and knowledge. They can't override its built-in guardrails.
Customize a welcome message
The welcome message is the first thing visitors see when they open the Assistant with no active conversation. Use it to set expectations or point users towards a specific workflow.
Add suggested questions
Suggested questions are pre-written prompts shown when the Assistant opens with no active conversation. They help users understand what they can ask, and can help you point your users towards useful answers or workflows.
Best practices for suggested questions:
Start with a real user goal (setup, troubleshoot, integrate).
Use the words your users use (avoid internal codenames).
Keep them specific. “How do I…?” beats “Tell me about…”.
Cover different intents: quickstart, how-to, troubleshooting, and reference.
If you’re embedding the Assistant in your product, you can also dynamically set suggestions in your embed configuration. See Customizing the Embed.
Choose where the Assistant appears
Under Appearance, set Placement to choose where readers reach the Assistant:
Sidebar (default) — the full chat experience, available on Ultimate site plans
Search box — inline answers alongside keyword search results, available on Premium and Ultimate site plans
The Assistant greeting applies to the sidebar placement only, so it's hidden while Placement is set to Search box.
Turning AI Assistant off and on again resets Placement to Sidebar. If your site uses the search box placement, set it again after re-enabling the Assistant.
Using AI Assistant in published docs
How readers reach the Assistant depends on its placement.
Sidebar placement
Users can access GitBook Assistant in three ways:
Press ⌘ + I on Mac or Ctrl + I on PC
Click the GitBook Assistant button next to the Ask or search… bar
Type a question into the Ask or search… bar and choose the 'Ask…' option at the top of the menu
Search box placement
With Placement set to Search box, users can ask a question directly in the Ask or search… bar at the top of the page, alongside the keyword results.
They can open this by clicking it directly, or by pressing ⌘ + K on a Mac or Ctrl + K on a PC.
As well as a summarized answer, below your users will also see an expandable section that shows the sources that the Assistant used to create its answer, plus related questions you can click as a follow-up.
The Assistant answers from the content of your docs site, and from any sources you've added through connections.
The Assistant does not answer across individual published sections on different docs sites.
Multi-section search is only available when viewing published sections that live within the same site.
Searching published documentation
Keyword search is automatically enabled on every site, whether or not the Assistant is turned on.
Your users can search for keywords within your docs site and jump quickly to specific pages or page sections across your entire site.
Embed AI Assistant in your product
You can embed GitBook Assistant directly into your product or website, giving users instant access to AI-powered help without leaving your application. The Assistant can be embedded as part of Docs Embed, which includes the Assistant tab for AI-powered chat, the Search tab for scoped discovery, and the Docs tab for browsing your documentation.
Choose the embedding method that fits your stack:
Standalone script tag – Quick setup with a
<script>tagNode.js/NPM – Server-side or build-time integration
React component – Prebuilt React components
Additional Assistant embedding guides:
Using embedded Assistant with authenticated docs – Required if your docs need sign-in
Customizing the Assistant embed – Welcome messages, actions, and suggestions
Creating custom embed tools – Connect Assistant to your APIs
API reference – All available methods and events
Extend AI Assistant’s knowledge
AI Assistant can use external knowledge through connections and MCP servers.
You can use connections when you want GitBook to sync records into your site, or use MCP servers when you want to hook up AI Assistant to custom tools.
Connections
Connections work best for content-heavy sources:
GitHub issues and discussions
Slack or Discord conversations
Support content
External docs, help centers, and websites
MCP servers
MCP servers work best for live tools and data:
Current account or product state
Internal systems that change often
Actions like creating tickets or filing bugs
Sources you don't want to sync into GitBook
Open your site's settings
Open your site dashboard. Then choose Settings → Connections.
Connect a source
Choose a source type. Then authorize it, or enter the URL to import.
Expose records to AI Assistant
Turn on Expose in search / assistant for the connection.
GitBook can then use those records in search and AI Assistant.
Open your site's settings
Open your site dashboard. Then choose Settings → AI & MCP.
Add a new server
In the MCP servers table, click Add MCP server.
Enter the server details
Give the server a name. Add its URL.
Then configure any HTTP headers GitBook must send with each request.
Last updated
Was this helpful?