Agent Gateway for cross-cloud extends Gemini Enterprise Agent Platform governance to any Cloud Native Computing Foundation (CNCF)-conformant Kubernetes cluster outside of Google Kubernetes Engine (GKE). Examples include on-premises Red Hat OpenShift clusters, clusters in bare-metal environments, and clusters in third-party clouds like Amazon Elastic Kubernetes Service (Amazon EKS) and Azure Kubernetes Service (AKS).
Agent Gateway for cross-cloud uses a decoupled three-plane architecture:
- SaaS management plane: A Google Cloud-hosted central plane for Agent Registry, Model Armor, Identity-Aware Proxy (IAP), Workload Identity Federation, and Google Cloud Observability.
- Cross-cloud control plane: A globally co-owned xDS control plane and an
in-cluster controller (
agent-governance-cross-cloud-controller). The controller watches services, workloads, and custom resources in your cluster and synchronizes them with Google Cloud for agent registration and gateway configuration. - Satellite data plane: Dedicated Envoy proxies (Agent Gateway for cross-cloud proxies) that you manage in your Kubernetes namespaces. Each proxy runs with an Agent Identity Runtime (AIR) sidecar that brokers workload identities and certificates for the proxy.
Shared responsibility model
Agent Gateway for cross-cloud integrates with your existing network stack without requiring you to replace your current infrastructure. This approach provides flexibility while maintaining centralized governance.
- Google's responsibility: Google manages the SaaS management plane and the cross-cloud control plane routing configuration. Google also provides standard Helm charts for deploying the Agent Gateway for cross-cloud proxies and satellite components.
Your responsibility: You are responsible for your cluster and network configuration outside of Google Cloud, including the following tasks:
- Forwarding traffic into the Agent Gateway for cross-cloud proxies using your existing service mesh or network stack.
- Injecting Kubernetes Service Account (KSA) tokens for identity federation.
- Ensuring outbound network connectivity from in-cluster control plane and data plane components to Google Cloud APIs. For details, see Allow outbound connectivity to Google APIs.
- Protecting your Google Cloud project and registered cluster resources from accidental deletion.
Identity-first zero trust
Agent Gateway for cross-cloud enforces identity-first zero-trust policies. You define Model Armor and IAP policies centrally in Google Cloud, and the proxy in your cluster enforces them for each request.
Pricing
During Limited Preview, Agent Gateway for cross-cloud is available at no charge. However, the other Google Cloud products that you configure with Agent Gateway for cross-cloud, such as Model Armor, Cloud Logging, and Cloud Trace, are billed at their standard rates. For details, see the following pricing pages:
Known issues
Agent Gateway for cross-cloud has the following known issues during Limited Preview:
- Fail-open behavior: The
failOpen: truesetting on extension bindings doesn't apply when the proxy can't reach Google Cloud, or when a binding references a Model Armor template that doesn't exist. For details, see Configure Model Armor. - Gateway readiness: A gateway can report that it's ready up to two
minutes before it accepts connections on port
8443. For details, see Create the gateway. - Agent Registry URL uniqueness: In-cluster service URLs must be unique across all clusters that are registered to the same project. For details, see Label workloads for auto-registration.
What's next
- Set up Agent Gateway for cross-cloud
- Deploy and register cross-cloud agents
- Enforce cross-cloud governance policies
- Configure cross-cloud agent observability