GitBook Assistant

GitBook Assistant gives users accurate, contextual answers drawn from your entire knowledge base — not just your docs

This feature is available on the Ultimate site plan.

The GitBook Assistant

GitBook 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 is available in the sidebar 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:

Enable GitBook Assistant

To enable GitBook Assistant, open your site's dashboard, navigate to the Settings page and choose AI & MCP from the menu on the left. Here you can enable GitBook Assistant from the options available.

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.

You can add suggested questions in your site’s Settings, under the AI & MCP section.

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.

Using GitBook Assistant in published docs

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

Embed GitBook 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 both the Assistant tab for AI-powered chat and a Docs tab for browsing your documentation.

Choose the embedding method that fits your stack:

• Standalone script tag – Quick setup with a <script> tag

• Node.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 GitBook Assistant’s knowledge

GitBook 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 GitBook Assistant to custom tools.

Connections

Best for

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

Best for

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

Add a connection

1

Open your site's settings

Open your site dashboard. Then choose Settings → Connections.

2

Connect a source

Choose a source type. Then authorize it, or enter the URL to import.

3

Expose records to GitBook Assistant

Turn on Expose in search / assistant for the connection.

GitBook can then use those records in search and GitBook Assistant.

Add an MCP server