Connectors

Connectors sync documents from external services into a knowledge base. New content is indexed, changed content is reprocessed, and content removed from the selected source scope leaves search after reconciliation.

This guide covers indexed knowledge bases. Organization Search has its own live-provider catalog and queries supported providers on demand; adding a knowledge-base connector does not enable it there.

Choose a connection method

Creating a connection requires workspace write access. For supported providers, Connection method controls who can search imported documents:

The available methods depend on the provider and deployment configuration. A workspace Admin must configure permission-aware methods or switch a connection out of Workspace mode. The Search source guides explain provider-specific permission setup.

For GitLab in Workspace mode, a PAT with read_api and access to the selected project is enough. The administrator-token and non-admin CSV permission paths apply to its permission-aware method; see GitLab.

Create a connection

Open a knowledge base, choose New connector, and select the service.

Manage a connection

Connected Sources shows the connected service, status, document count, and available last/next sync details. Open Connection actions for:

After a healthy sync completes with no failed documents, a one-minute cooldown applies before another manual sync. Error recovery and failed runs bypass that cooldown.

Before removing a connection, review its document consequence. In Workspace mode, Also delete synced documents is optional; otherwise the documents remain as plain documents. Permission-aware connections require deletion of their synced documents when removed. Logical removal can finish before physical storage cleanup completes.

Sync history

History shows the latest 10 runs, with time, state, document-change counts, and error or notice details. Counts identify documents added, updated, deleted, failed, or skipped; No changes means no content changes were recorded.

In progress… means work is running. Continuing can mean a large listing is resuming across runs rather than failing. Partial, Failed, or Interrupted entries need their accompanying notice or error inspected. Member-account connections also report member sync progress and accounts that need reconnecting.

Exclude documents

1. Open Settings for the connection and select Documents.
2. Use the document-status picker to inspect Included, Excluded, Failed, or Skipped documents.
3. Select Exclude on a document to remove it from search and skip it on later syncs, even if the source changes.

To reverse this, select Excluded and choose Restore. Retained completed content can become available again; the next sync refreshes it from the source. For failed or skipped documents, read the reason and fix the source or configuration. Retry failed documents when offered, or sync again after correcting skipped source content.

Sync behavior and recovery

A sync lists the selected source scope and compares it with stored content. New documents are processed; changed documents are re-chunked and re-embedded. Unchanged content usually avoids re-indexing, although the source may still be listed or read to verify changes and permissions.

A base can have multiple connections. Each owns its documents, and a Knowledge block can search their combined content within the caller's access. Their metadata definitions share the same tag slots.

Content removed from the source scope

Content deleted or removed from the selected source view leaves search after a successful reconciliation. Examples include:

• Archived Confluence pages or Webflow CMS items.
• Asana tasks whose projects are all archived; a task still reachable through an active project remains included.
• Retired ServiceNow articles when no explicit workflow-state selection overrides the default.
• Files in Google Drive trash or Outlook messages in Deleted Items.
• Canceled incident.io incidents under the default filter; All (including canceled) or an explicit selection can include them.

An incomplete source listing keeps unlisted documents instead of treating them all as deleted. Incremental listings also do not prove that an unlisted item was removed. Restoring content at the source makes it eligible for a later sync; its document ID is not guaranteed to change.

Statuses

A single-document failure can leave the rest of the sync running. Transient failures can be retried; unsupported or invalid content needs a source/configuration change. Read the Documents filters and history to distinguish failed indexing from a failed source listing.

Full-sync failures increase the retry delay. After 10 consecutive failures, the connection is disabled. Correct the source or credential issue, reconnect an invalid OAuth grant when needed, and choose Resume syncing. Resume alone does not repair a revoked credential.

Confluence attachments

Matching pages and blog posts include PDF, Word (.docx and Word 97–2003 .doc), Excel (.xlsx), and PowerPoint (.pptx) attachments as separate documents. Attachment indexing requires read:attachment:confluence; include it in a service account's scoped token. See Confluence setup.

Attachments are checked even when the parent page is unchanged. Files over 100 MiB appear as skipped. Convert Word 6/95 files to .docx, .xls to .xlsx, and .ppt to .pptx. Existing connections pick up supported attachments on later syncs.

Image attachments

Image Attachments → Read with AI converts PNG, JPEG, WebP, and GIF attachments into searchable text. GIF extraction uses the first frame. Tables become rows with headers, scans become extracted text, and diagrams or screenshots include visible text and a description. Decorative images remain documents with zero chunks.

Images smaller than 2 KiB, SVG, HEIC, and AVIF files are not indexed through this image path. It needs a Gemini key: an available workspace key first, otherwise the deployment's key. Without a key, images are deferred until one is configured.

New knowledge-base Confluence connections enable image reading by default. Connections created before image support keep it off until you choose Read with AI; their next sync then indexes images.

The default admission limit is 500 new or changed images per sync. Remaining images wait for another sync. With Member accounts, each run that continues a sync can read up to 500. Self-hosted operators can configure KB_IMAGE_EXTRACTIONS_PER_SYNC. Unchanged images reuse their extraction, including during a full resync; transient or malformed responses can cause additional model requests.

Image extraction follows model billing: BYOK usage is billed by the provider, while platform-key usage follows the deployment's Sim billing policy. Choose Don't index to stop image indexing; the next sync removes indexed images. Switching back processes them again and incurs the applicable model usage.

Available connectors

The source picker lists the available services. Built-in connectors include:

Related docs

• Tags and filtering — definitions, slot capacity, and operators.
• Debugging retrieval — verify indexed content and access.
• Using a knowledge base in a workflow — retrieve synced passages.
• Organization Search — live providers and permission-aware source setup.