One control plane. Your infrastructure. Three layers that actually connect.
Most multi-cluster diagrams are a pile of logos. This is the system: how clusters join, how data moves, and how names fail over — without a proprietary runtime.
Clusters you own
EKS in your AWS account, GKE in your GCP project, k3s on Proxmox or your own hardware. Zeus provisions and operates them; it does not host them. Standard Kubernetes underneath — kubeconfig exportable anytime.
Global fabric
Encrypted overlay across every enrolled cluster. Non-colliding pod and service CIDRs from Network Plans. Default-deny grants (directional, revocable). Continuous reachability probes. Mesh CA trust injected into pods so TLS just works cross-cluster.
Global data + DNS
Multi-region databases with a shared ops surface (topology, lag, backups, failover controls). Health-checked DNS for mesh names and public multi-region publish to Route53 and Cloud DNS — same health model, two planes.
From connect → fabric → data → name.
- 1 Connect
Link AWS/GCP or connect Proxmox hardware. Nothing inbound required on your side.
- 2 Provision
Same wizard shape for EKS, GKE, and Proxmox/k3s. Opinionated defaults; Advanced knobs when you want them.
- 3 Enroll fabric
Non-colliding CIDRs, encrypted tunnels, explicit grants. Fresh clusters can reach nothing until you allow it.
- 4 Deploy
One service definition → any cluster. Environments layer overrides. No kubectl on the main path.
- 5 Replicate
MySQL, PostgreSQL, ClickHouse, Yugabyte — global when you say so. Engine-honest recovery paths.
- 6 Publish names
Mesh DNS for private service discovery; public DNS for customer-facing hostnames. Health drives both.
Clear boundaries keep the product honest.
Not a managed Kubernetes host
Clusters stay in your accounts and racks. Zeus is the control plane in front of them.
Not Cilium ClusterMesh
Any reasonable in-cluster CNI. Cross-cluster joining is the fabric layer — not a CNI monoculture.
Not a black-box PaaS
Advanced shows real settings and generated manifests. Leave and the clusters keep running.
Not identical physics for every database
One ops board for every engine. Recovery matches how MySQL, Postgres, ClickHouse, and Yugabyte actually work.
Specificity only a real product can claim.
- ✓ Continuous mesh reachability probes — not a one-time setup script
- ✓ Default-deny grants with a map that matches enforcement
- ✓ Mesh + public DNS on one health brain
- ✓ No kubectl required to ship; Advanced never lies
- ✓ EKS · GKE · Proxmox/k3s on one workflow
Walk the architecture on live infrastructure.
Book a demo of fabric, global data, and failover — or start free on your own accounts.