Fourteen connector types read content over each vendor's REST API. They are
for content meant for everyone who uses the assistant: published help articles,
public site pages, one course's pages, one repository's issues.
These sources follow each vendor's published API and are tested against a local
server that returns the vendor's response shape. They have not yet been verified
against live vendor accounts. Run a first sync on a small scope and check the
run before relying on one.
Whole workspace, no per-person sharing. A source indexes whatever its credentials can read and makes it available to everyone who can use the project. It does not mirror who can see what at the vendor. Give it credentials that can read only what everyone may see.
Content the vendor limits is left out. Where a vendor marks an article as limited to signed-in users, agents or a segment, the knowledge base sources leave it out and report it on the run as restricted_at_source. Zendesk and GitBook have an admin opt-in to include it.
Every sync lists the whole source. Unchanged documents are skipped, and a document that is no longer listed is deleted from the index.
A refused sign-in fails the run with source_refused; an unreachable source fails it with source_unreachable.
A record with an unexpected shape is reported on the run as invalid_document and left out. An empty record is left out.
They are created like any other connector: see Connect a source. Field names below are the keys of credentials and settings.
subdomain (the part before .zendesk.com), include_restricted_articles (default false)
Indexes
Help centre articles. Drafts are left out. Articles limited to agents or a user segment are left out and reported as restricted_at_source unless include_restricted_articles is on, which makes them available to everyone who can search the collection, including a public widget
Published articles (status 2) in folders visible to all users. An article in a folder, or under a parent folder, limited to signed-in users, agents, companies, bots or segments is left out and reported. Drafts are left out
Answers link to https://<domain>.freshdesk.com/support/solutions/articles/<id>.
client_id, client_secret, refresh_token from a Self Client in the Zoho API console, with the Desk.articles.READ scope
Settings
org_id, data_centre (com by default; also eu, in, com.au, jp, ca, sa, com.cn)
Indexes
Published articles whose permission is everyone (ALL). Articles for registered users or agents are never read
Zoho access tokens last an hour, so every sync exchanges the refresh token for
a new one. Each sync lists every article (three API credits per 50) and reads
only the bodies changed since the last sync (one credit each).
project_version_id (empty reads every public knowledge base workspace), language (default en), site_url (used for links when Document360 returns only a path), api_host (only for the US data centre, https://apihub.us.document360.io, or private hosting)
Indexes
The published version of each article. Hidden articles and articles limited to signed-in readers are left out and reported; drafts are never read
Every document page of one space, as Markdown. Links and generated pages are left out
A space that is not public (private, unlisted, share link, visitor
authentication) is left out and reported, and pages indexed while it was
public are removed, unless include_non_public_space is on. Only pages edited
since the last sync are read again.
Tickets, Slack messages, CRM records and GitHub pull requests are covered by
Workplace tools. Airtable attachments and
anything that needs per-person permissions are outside these sources. If you need a
source that is not here, see Custom connector.