Skip to main content

Knowledge Bases and Connectors

Knowledge Bases and Connectors describe different parts of the enterprise knowledge path. A Knowledge Base points to a searchable index. A Connector moves source documents into customer-owned storage that can feed that index.

Customer-hosted Connector data flow and Taimoe Knowledge Base control plane

Knowledge Base​

A Knowledge Base is a Team-owned registry entry for an external retrieval source. It stores the source type, provider-specific index configuration, a cloud credential reference, status and freshness metadata, and optional lineage to the Connector that feeds it.

An Agent-to-Knowledge-Base binding states which registered sources an Agent may use. Vertex AI Search is the current implemented source. Other source types in schemas or design documents are future compatibility points unless explicitly documented as available.

Cloud credential​

A Team Cloud Credential supplies the identity used by the platform-mediated Knowledge Base path. A Knowledge Base references the credential instead of copying credentials into every Agent.

Service-account JSON is currently stored in the platform database as JSON; application-level encryption at rest remains a production hardening requirement. Application Default Credentials avoid storing a secret in the row but depend on the deployment identity being configured correctly.

Connector​

A Connector is a separately deployed sync service. The current SharePoint Connector runs in the customer's environment and moves documents from SharePoint to customer-owned object storage. Taimoe stores configuration versions, validation and publication state, health, and sync-run metadata; it does not receive the documents or SharePoint credentials.

Connector configuration is append-only:

edit -> save draft -> validate on the Connector -> publish -> supersede

The control token protects calls from Taimoe to the Connector. A separate ingest token authenticates sync reports from the Connector to Taimoe. The ingest token is shown once and only its hash is stored.

Governance boundary​

Platform-mediated retrieval can use Team ownership, managed credentials, and audit metadata. A Runtime that directly calls an index with its own credential bypasses those application-level controls. Provider IAM and network policy are required when direct access must be prohibited.

Knowledge Base management and the SharePoint Connector control plane are available. Platform-managed Connector scheduling is planned; customer-side scheduling remains authoritative.