Connections

Connect external sources to GitBook Assistant and GitBook Agent

Connections let you bring external content into your docs site.

Connected content can appear in AI Assistant, so visitors get answers drawn from everything you publish rather than your docs alone. Some connections also help GitBook generate change requests and identify content gaps. An MCP server connection can additionally let visitors carry out tasks from your docs.

In your site’s sidebar, go to Settings → Connections.

Available connections

You can connect support platforms, developer communities, and other content sources:

• Intercom, Zendesk, Freshdesk, Front, HubSpot, Zoho Desk, and Pylon

• GitHub Discussions and GitHub Issues

• YouTube, a website, or an MCP server

How connections work

1

Connect a source

Select a source type. Then authorize it, or enter the URL you want GitBook to import.

If you connect YouTube, enter the channel ID — not the channel name.

A channel name won’t work for this source.

In YouTube Studio, open Settings → Channel → Advanced settings.

Copy the Channel ID.

If your channel URL contains /channel/, copy the part after /channel/.

Each connection creates a stream of records from that source, such as issues, discussions, tickets, conversations, pages, or videos.

2

GitBook syncs records

After you save the connection, GitBook starts syncing records from that source.

In the connections list, you can review the sync status, record count, and last sync time.

Connected URLs refresh once a day by default. You can also trigger a refresh manually from your site’s Settings → Connections.

3

Choose how the records are used

For each connection, you can choose whether its records appear in search and GitBook Assistant, and whether they help GitBook generate change requests.

You can also adjust the connection’s search ranking to prioritize or deprioritize its records.

Reviewing your connections

The Connections page has three parts: Connect a source to add a connector, Connections to manage the ones you already have, and Records to browse what GitBook ingested.

The connections table

Every connected source has a row:

Column	What it shows
Source	The connector, and the account, channel, or URL it points at.
Status	Whether the most recent sync finished.
Records	How many items the source contributed, counted in that source’s own units — pages, videos, discussions, or conversations.
Search	The connection’s search ranking boost. This is empty when Expose in search / assistant is off.
Last synced	How long ago the connection last refreshed.

Records

The Records list shows the individual items GitBook ingested across every connection, newest first. Each record gives its title, the connector it came from, when GitBook processed it, and a link to the original.

A large site accumulates thousands of records, so the list is paginated. Use it to confirm a source brings in what you expect before you turn on Expose in search / assistant: a website connection returning blog posts when you wanted product pages shows up here first.

Connection settings

When you edit a connection, you can configure these options:

Label

Use Label to give the connection a clear name.

This makes the source easier to identify in the connections list, especially if you connect multiple repositories, websites, or channels.

Auto-generate change requests

Auto-generating change requests is in beta.

See Content gaps for how GitBook identifies gaps in your docs.

Turn on Auto-generate change requests to let GitBook learn from the connection’s records and suggest documentation updates.

GitBook can use those records to create change requests with suggested changes for your team to review.

Expose in search / assistant

Turn on Expose in search / assistant to make records from this connection searchable by visitors and available to answer questions.

This allows connected content to appear in your site’s AI-powered search experience and in GitBook Assistant responses.

Enabling Expose in search / assistant makes records from that connection available to anyone who can access the site — including external visitors if the site is shared externally.

Search ranking boost

Use Search ranking boost to prioritize or deprioritize records from a connection in search.

Increase the value if this source should appear more often. Lower it if this source should have less weight than your primary docs.

Connector details

All connectors in this section can provide context to AI Assistant. GitBook Agent can use resolved records to identify content gaps and suggest documentation updates.

Intercom

GitBook indexes resolved conversations with at least one reply from a human or AI support agent. It doesn’t index open, unresolved, or spam conversations.

To connect Intercom, use an API key with read access to conversations. You control its scope and expiry.

Zendesk

GitBook indexes resolved conversations with at least one reply from a human or AI support agent. It doesn’t index open, unresolved, or spam conversations.

To connect Zendesk, enter your Zendesk subdomain, such as gitbook.zendesk.com. You also need an API key with read access to conversations. You control its scope and expiry.

Pylon

GitBook indexes resolved conversations with at least one reply from a human or AI support agent. It doesn’t index open, unresolved, or spam conversations.

To connect Pylon, use credentials with read access to conversations.

GitHub Discussions

GitBook indexes resolved discussions with at least one reply. It doesn’t index unresolved or spam discussions.

To connect GitHub Discussions, use credentials with read access to discussions.

GitHub Issues

GitBook indexes resolved issues. It doesn’t index open, unresolved, or spam conversations.

To connect GitHub Issues, use credentials with read access to issues.

Data stored for these connectors

GitBook stores only the conversation, discussion, or issue. We don’t store metadata or personal information.

GitBook redacts sensitive information before storage. We don’t store sensitive information in the indexed record.

Data retention and privacy

What does GitBook index?

GitBook indexes content from your connections and uses it across search, change requests, and other content-driven features. Each item we index from a connection is called a record. Examples of records are Intercom tickets, Freshdesk tickets, website pages, and more.

We use records to suggest content changes and provide additional context to agents running across GitBook.

We retain records for as long as the source remains connected. If you remove the connection, we delete all stored records. We do not delete any content that was generated using those records as context, or remove the content of records from any stored AI conversation logs.

For example, you might connect Intercom and GitBook suggests a documentation change based on a closed Intercom support ticket. If you then remove the Intercom connection, all our Intercom specific stored data is deleted, but the change requests will remain.

FAQ

What is your policy around PII?

Generally, GitBook only requests source fields that it can use as context (see connector pages for more specifics on each source).

When we ingest a record, we pass it through PII redaction processes. The first is an in-memory redaction of any emails, phone numbers, addresses, and other personal information.

We then pass the record through a 2nd PII step using a 3rd party LLM. This is a ZDR-compliant, no-store request that is not stored on the servers of our 3rd party and is not used in any training.

PII may exist on our servers for a few minutes whilst we do this pre-processing, but we only store the post-processed record.

Do you send data to any 3rd parties?

We use a 3rd party to redact PII after our initial, internal PII redaction is done.

We also use the redacted record content in 3rd party providers when creating change requests and responding to user questions.

For example, when writing content with GitBook Agent, the agent may include content from an external record as additional context.

Does GitBook use data for any model training, either itself or via 3rd parties

No, we do not use record data for training and verify with our 3rd party providers that they do not either.

Can I limit which records GitBook fetches?

Connector-specific requirements and indexed records are listed in Connector details.