This shows a (management) Kubernetes cluster with Cluster API, Metal3 and Kamaji deployed. There are two workload clusters. One is a single-node cluster using the common kubeadm control-plane provider. The other cluster has a hosted control-plane, meaning that the control-plane is managed by Kamaji and running as pods in the management cluster. A worker node (using Metal3) is joined to this control-plane, creating a cluster with only a single worker node (and no control-plane nodes).
Kamaji runs the control-plane as two pods in the default namespace. Each pod contains all the control-plane components, except etcd. Etcd runs separately in the kamaji-system namespace.