Four steps from connected infrastructure to multi-region operations.
Intent-first defaults, Advanced when you want the knobs, standard Kubernetes underneath. No kubectl required along the path.
Connect your infrastructure
Link an AWS or GCP account, or connect Proxmox and your own hardware. Nothing inbound to expose. That gear becomes another place Zeus can put a cluster.
Provision a cluster, the same way every time
Pick where it lives and what it is for. Zeus chooses sane CNI, node groups, encryption, and topology. The flow is the same on EKS, GKE, and Proxmox/k3s. Every cluster lands in one list.
Join clusters into one fabric
Link two or more clusters. Zeus stands up the encrypted cross-cloud overlay — non-colliding IP space, default-deny grants, continuous reachability probes, service discovery by name. Pods get mesh CA trust injected so apps do not each special-case TLS.
Global data and health-checked DNS
Deploy MySQL, PostgreSQL, ClickHouse, or Yugabyte and mark them global. Publish mesh and public names that follow health. Apps keep normal endpoints; geography stops being a custom project.
The full surface, not the highlight reel.
Working control plane, verified on real multi-cloud fleets — not a thin wrapper around one script. Here is what is in the box.
Provisioning
- Multi-cloud, multi-accountAWS and GCP with per-cluster account binding, plus bare-metal and Proxmox/k3s on your own hardware. One console, every cluster you own.
- Intent-first wizardsYou say what you want to run; ZeusK8s picks sane CNI, node groups, encryption, and topology. The knobs are still there under Advanced.
- Node groups & autoscalingEKS-style managed node groups on every flavor, including k3s, with labels, taints, and live scale up/down over the agent tunnel.
- Clean teardownDestroys sweep tagged resources with a keep/delete checklist (EBS, KMS, SGs, load balancers) instead of leaving orphans and a surprise bill.
The global fabric
- Cross-cluster service discoveryServices resolve and connect across clusters and clouds by name, over an encrypted overlay, routed to real pod IPs with no NAT. Continuous probes verify the path stays healthy.
- Default-deny by designA freshly connected cluster can reach nothing and is reachable by nothing. Every cross-cluster flow is an explicit, directional grant you can revoke in a click.
- WAN-aware ingressPin egress IPs, route ingress to the nearest region, and bring traffic in on a stable edge, all without hand-editing a load balancer per cluster.
Stateful data
- Global database replicationMulti-region MySQL, PostgreSQL, and ClickHouse at deploy time. Shared ops board for topology, lag, and actions. Apps keep a normal endpoint.
- Engine-honest failoverMySQL multi-cluster auto-arbiter when eligible; guided switchover and rebuild for every engine. One surface — recovery physics that match reality.
- Backups & point-in-time restoreScheduled backups to S3-compatible storage and PITR restores, including standing up a fresh cluster from a backup.
Day-two operations
- Guided upgradesEnd-of-life banners, preflight checks, and a stepped control-plane → addon → node-group upgrade flow. No more cluster-version roulette.
- Live pods, logs, and metricsLens-grade workload views, multi-pod log tailing, and metrics pulled straight from Prometheus, without leaving the tool.
- Infra add-ons as first-classHelm-based add-ons (ingress, cert-manager, databases, observability) tracked per cluster with a real config UI, not a values.yaml you paste and pray.
- Workload rightsizingLive CPU, memory, and throttling analysis with concrete request/limit recommendations — so you stop overpaying for idle cores and under-provisioning the hot path.
Want the four steps live, end to end?