Data Connectors
Data connectors keep an AI Employee's Knowledge base current by syncing documents from external sources — SharePoint sites, Google Drive folders, Confluence spaces, and Box folders. Where Tools let an AI Employee act in external systems, connectors let it know about your content: every synced document is extracted, chunked, embedded, and made searchable so the AI Employee can cite it in its answers.
Connectors are different from Tools. A Tool call is a sub-second request that runs through the integrations service. A connector sync is a long-running background job — listing files, downloading content, extracting text, embedding — so it runs in the ingestion service as a durable workflow with retries and incremental updates. You manage connectors from the Data sources tab of the Integrations page, and credentials are admin-only.
The connection model
Three concepts work together:
- Connection — holds the credentials and configuration for one external source (admin-only). One connection can feed multiple Knowledge bases scoped to different paths.
- Knowledge base (KB) — a searchable collection linked to a connection (and a sync schedule). KBs are scoped to a specific AI Employee.
- Sync — the job that pulls documents from the source into the KB. Syncs run on a schedule or can be triggered manually.
Roles. Creating connections and Knowledge bases is admin-only — builders never see credentials. Builders select from available KBs when configuring a workflow and can narrow results with a folder filter at search time.
Supported connectors
The Data sources tab shows a tile for each source. Local file upload is shown for reference but is handled in the AI Employee's Knowledge bases tab, not here. The external connectors are:
| Connector | Auth | What you connect at | Setup guide |
|---|---|---|---|
| SharePoint | App client-credentials (Azure AD app); or OAuth2 delegated (sign in with Microsoft) | A SharePoint site | SharePoint Connector |
| Google Drive | Service account, or OAuth2 user consent | A Drive folder | Google Drive Connector |
| Confluence | OAuth2 (Atlassian authorization code), or Basic auth (email + API token) | A Confluence space or pages | Confluence Connector |
| Box | OAuth2 (authorization code) | A Box folder | Box Connector |
Each connector connects at the collection level — a site, folder, or space — not the whole instance or tenant. When you connect a collection, Ema ingests its content, including nested folders and attachments, within that scope.
How documents flow in
Every connector feeds the same pipeline, so mixing sources never requires re-indexing:
- List — the connector enumerates the files in scope.
- Download — it fetches each file's content.
- Extract — text is extracted from PDFs, Office documents (
.docx,.xlsx,.pptx), HTML, and images (via a vision model). - Chunk and embed — text is split and embedded into vectors.
- Store — chunks land in the vector store and become searchable.
Syncs are incremental: Ema detects changes with the source's ETag, falling back to a content hash, so unchanged files are skipped. A background scheduler runs due syncs automatically; you can also trigger a sync on demand.
Setting up a connector
From the Data sources tab, select a connector tile and complete its setup form. Every connector asks for a name and description; the rest depends on the source. Each connector has a dedicated setup page with the full field list, scopes, sync configuration, and limits — the summaries below link to them.
SharePoint
SharePoint Online defaults to an app (client-credentials) mode using an Azure AD app registration, with an optional delegated OAuth2 mode where you sign in with your Microsoft account. App mode collects the site URL, Azure tenant ID, client ID, and client secret; delegated mode asks only for a name and site URL, then signs you in for the scopes Sites.Read.All, Files.Read.All, and offline_access. Full setup, scopes, and limits: SharePoint Connector.
Google Drive
Google Drive supports two modes — full setup, scopes, and limits: Google Drive Connector.
- Service account — paste the full service-account JSON; share the folders or Shared Drive with the service account.
- OAuth2 user consent — sign in and approve access; Ema uses platform OAuth credentials.
Confluence
Confluence Cloud supports the Atlassian OAuth2 authorization-code flow (default when configured) and Basic auth (account email + scoped API token). You scope a sync by Confluence space keys and/or root page IDs. Full setup, scopes, and limits: Confluence Connector.
Box
Box uses the OAuth2 authorization-code flow with platform-level Box app credentials (one Box app serves all tenants). Connecting opens Box's consent screen; each tenant's access and refresh tokens are stored encrypted per connection and refreshed at sync time. Full setup and limits: Box Connector.
OAuth2 connectors validate the connection during setup — the success banner confirms the account that authorized it. API-key and client-credential connectors run a live test call against the source before the connection is stored, so wrong credentials fail immediately rather than at first sync.
Connecting from the OAuth flow
OAuth2 connectors follow a consistent browser flow regardless of source:
- You enter a connection name and select Connect.
- Ema redirects you to the provider's sign-in and consent screen, carrying a signed state parameter for CSRF protection.
- You approve the requested scopes.
- The provider redirects back to Ema's callback, which exchanges the one-time code for tokens and stores them encrypted.
- The Data sources tab shows a success banner; the new connection appears in the connector's list.
If something goes wrong — an expired state, a token-exchange failure, or a failed test — the tab shows a specific error banner so you can correct it.
Grounding an AI Employee in connected content
Once a connection exists, an admin creates a Knowledge base from it (scoped to an AI Employee) and the documents in scope sync into that KB. A builder then references the KB from the relevant workflow node — for example a search-and-respond agent — and can apply a folder filter to narrow what that node searches. From then on, the AI Employee retrieves and cites the synced content in its answers.
What's next
- Per-connector setup guides: SharePoint Connector, Google Drive Connector, Confluence Connector, Box Connector.
- Integrations Hub — the full Integrations page, connections, and MCP servers.
- The Tool Editor — give an AI Employee the ability to act in external systems.