Deployment

Deploy Cerbos on Kubernetes

Run the Cerbos PDP on Kubernetes as a standalone service or sidecar container alongside your application pods.

Flexible topology

Deploy as a centralized service, a sidecar, or a DaemonSet depending on your latency and isolation requirements

Horizontal scaling

Cerbos is stateless and scales horizontally, add replicas to match your traffic without any coordination

Observable

Built-in Prometheus metrics, distributed tracing, and health check endpoints for liveness and readiness probes

What is Cerbos?

Cerbos is an open-source authorization layer that decouples access control from your application code. It runs as a stateless Policy Decision Point (PDP) that evaluates fine-grained policies at request time.

Authorization policies are written in human-readable YAML supporting RBAC, ABAC, and conditional rules. They can be updated, tested, and deployed independently of your application.

Deploying Cerbos via Kubernetes gives you a production-ready authorization service that scales horizontally and fits naturally into your existing infrastructure and observability stack.

Policy-as-codeHuman-readable YAML policies managed like source code Scalable PDPStateless policy decision point with sub-millisecond latency Centralized managementManage, test, and deploy policies from a single control plane

How to deploy Cerbos on Kubernetes

Choose a deployment topology, Decide between a standalone Deployment with a Service, a sidecar container in each pod, or a DaemonSet.
Deploy the Cerbos container, Use the official Cerbos container image or the Helm chart to deploy the PDP into your cluster.
Configure policy loading, Point Cerbos at a ConfigMap, Git repository, or Cerbos Hub bundle for policy storage.
Connect your services, Use a Cerbos SDK to send authorization checks from your application pods to the PDP.

FAQ

Should I deploy Cerbos as a service or a sidecar?

A standalone Deployment with a Service is the simplest approach and works well for most setups. Sidecar deployments eliminate network hops and are useful when you need the lowest possible latency or want to isolate authorization per pod.