Multi-tenancy: auth, isolation, and RBAC¶
Maps to: Multi-tenancy: authentication, authorization, Agent Auth, RBAC.
Scope¶
Isolating end-user data, letting agents act with user-scoped credentials on third-party systems, and controlling operator access to deploy and observe.
Design questions¶
- Identity propagation from edge API through to tool calls and trace metadata.
- Resource tagging and filter rules on reads versus writes.
- Separation between end-user tenancy and internal operator RBAC.
- OAuth token storage, refresh, and revocation when users disconnect integrations.
Tradeoffs¶
- Custom auth middleware maximizes flexibility; misconfiguration causes cross-tenant leaks.
- Agent Auth reduces credential sprawl but centralizes sensitive token storage.
- Strict RBAC can slow iteration for small teams; loose RBAC fails enterprise review.
Evaluation hooks¶
- Cross-tenant access attempts on threads, memories, and traces must fail closed.
- Agent calls third-party API with correct user scope only.
- Operator roles cannot perform actions outside assigned policies.
Reference notes¶
See LangChain runtime article (multi-tenancy auth layers figure).