KuberNation

Your Kubernetes cluster as an explorable world map.

Instead of scrolling tables of pods and nodes, you look at a map. Each node is a patch of terrain coloured by its health, each workload is a city sited on the node its pods run on, and the problems that need you surface in a queue — the next thing needing your attention — rather than buried in dashboards.

Download View on GitHub
A live Kubernetes cluster rendered as an isometric world map, with cities for workloads and an attention queue in the right column.
A live cluster as an isometric world — terrain is node health, cities are workloads, and the right-column attention queue ranks what needs you.

Why a map?

Table and dashboard tools hide two things a map makes obvious: where things run, and what matters right now. KuberNation is built around four ideas.

Spatial, not tabular

Resources project onto a stable world. A node's terrain is its health; a workload's city moves when its pods reschedule; two clusters sit side by side so drift is visible at a glance.

Attention-driven

Crash loops, stalled rollouts, pending volumes, nodes under pressure, burning error budgets — aggregated and ranked into one queue. Press N to fly to the next problem.

Read-first writes

It observes by default. Every change is explicit, confirmed, RBAC-checked, and server-side dry-run validated — and all the write code lives in one small, auditable file.

Ask a model

Consult a local or corporate LLM to explain a workload or the whole realm — from a redacted, fenced summary, never raw dumps or Secrets. The model proposes; you and the gate dispose.

A whole cluster, as geography

Everything on screen is a real Kubernetes object you could kubectl get — the game words are just a lens.

Drill-downs

Cities and provinces

Click a city to open the workload in full; click open land for the node. Both are classic 4X-style screens reframed for Kubernetes.

  • City window — replica/update gauges, a pod census, owned Services, Ingresses & PVCs, the availability treasury, rollout history with one-click rollback, and a merged change log.
  • Province window — zone & health, CPU/memory gauges with live trend sparklines, the garrison of pods, terrain facts, and node conditions.
The city window for a Deployment, showing gauges, pod census, improvements, treasury, and change annals.

Map views

Seven ways to colour the board

The View menu recolours the whole world like a strategy game's map modes, so a different question gets a different map.

  • Terrain (health) · Pressure (CPU/mem) · Replicas (workload strength) · Namespace (territory).
  • Saturation — fullness toward hard limits, catching the silent max-pods failure Pressure misses.
  • Walls — NetworkPolicy segmentation gaps · Upkeep — what each node costs to run.
The map recoloured as a namespace territory map.

Triage

What's wrong, what else it hits, what changed

Three axes answer an incident, and each has a home.

  • Attention queue — ranked problems with one-line next-action hints.
  • Blast radius (B) — the dependency fan-out across the map, with a clickable impact list (honest topology, never invented edges).
  • The Annals (H) — a classified change-feed with a "trouble begins here" fault line, plus a one-click after-action report.
The Annals change timeline with a fault line marking the first failure.

Advisors & cost

Read-only reports for the whole realm

Seven advisor tabs roll up the cluster as pure functions of what's observed.

  • Health · Storage · Network, plus Right-sizing (requests vs. live usage, with reclaimable totals).
  • Cost cartography — an Upkeep overlay + advisor; honest "cost units" by default, real $/mo with your rates, or invoice-grade via OpenCost.
The Right-sizing advisor showing over- and under-provisioned workloads and reclaimable CPU.

Security & posture

See your exposure, before an attacker does

A small security suite, all read-only and all reusing what KuberNation already watches.

  • The Charter — a self-scoped can-i grid of what you can do here, dangerous capabilities flagged. It kills surprise 403s.
  • Hardening scan (PSS / OWASP) · Walls (segmentation, OWASP K07) · a 0–100 posture score.
The Charter: a can-i grid of the operator's permissions with dangerous ones highlighted.

Game Day

Break it on purpose, watch it heal

A chaos-engineering console: inject a real, confirmed, RBAC-gated failure and watch the cluster respond — the queue lights up, the blast radius spreads, the budget spends.

  • Nine experiments (kill pods, outage, scale spike, broken image, node failure, partition…) or a compound Skirmish / Raid / Siege drill.
  • A scorecard grades recovery and time-to-detect; control-plane & system namespaces are refused, and drills auto-restore so the cluster is never stranded.
The Game Day chaos console with experiment choices and a blast-radius preview.

The Oracle — bring your own model

Consult a language model you bring to explain a workload, a node, or the whole realm. Advisory, private by construction, and it can never touch the cluster on its own.

  • It sees a redacted, fenced summary, never your cluster — built from what KuberNation already observed, with a mandatory preview of the exact bytes before anything is sent.
  • Bring your own endpoint — a local model (Ollama by default) or any OpenAI-compatible / corporate endpoint. Remote egress stays behind an explicit per-session arm gate.
  • It streams its answer, with one-click links to drill into the worst concern and lenses to fold in logs, storage, or blast radius.
  • It can propose a fix, never apply one — a suggestion enters the planning turn and commits only through the same dry-run + RBAC gate.
The Oracle modal showing the mandatory pre-send preview of the exact redacted, fenced payload.
The mandatory pre-send preview — exactly what leaves your laptop, with credential redaction and a prompt-injection fence.

The mapping

Every shape on the map is a live Kubernetes object.

Kubernetes objects and how each appears on the map
KubernetesOn the map
Zone / failure domaina continent — nodes in the same zone cluster together
Nodea province of terrain, coloured by health
Workload (Deployment, StatefulSet)a city, sited where its pods run — it moves when they do
Poda citizen of its city / garrison of its node
DaemonSeta road, paved across every node it runs on
Service / Ingressa harbor / gate on the city's coast
PersistentVolumeClaima granary inland of the city
Job / CronJob / custom resourcestructures on the namespace islands
A problema ranked attention-queue entry + a flag on the map

Honest about what it is

KuberNation is an operator-laptop tool. It talks to your cluster through your kubeconfig and runs no in-cluster agent. The whole write surface is one auditable file; there is deliberately no exec/shell, and Secret contents are never surfaced. The only call KuberNation makes to a third-party endpoint is the optional Oracle — opt-in, off by default for remote, and gated like everything else (OpenCost, if used, is read through your existing cluster connection).

Install

Pre-built binaries are attached to every GitHub release. Download, verify against SHA256SUMS, and run it against your current kube-context. From source: cargo run --release -- --context <ctx>.

macOS

Universal .dmg

Apple Silicon + Intel. Signed & notarized — open the .dmg, drag KuberNation to Applications, and launch. No Gatekeeper workaround needed.

Linux

x86_64

Unpack and run against your current context:

tar xzf kubernation-*-linux-x86_64.tar.gz ./kubernation
Windows

x86_64

Unzip and run. Unsigned, so SmartScreen may warn on first launch — choose “More info ▸ Run anyway”.

kubernation.exe

It's a windowed desktop app (not a TUI), so it needs a display — it runs on your laptop, not over SSH. It needs read access to the usual kinds; the few write actions each check RBAC and show as locked if you lack the verb.