Quickstart

Try the whole ladder
without touching a GPU.

The end-to-end test posts a synthetic XID 79, parks the incident at the reboot rung awaiting approval, approves it over the API, and asserts the full audit trail through to resolution. No cluster, no hardware, about fifteen minutes.

  • Kubernetes 1.29+ for the operator path
  • NVIDIA GPU Operator supplies dcgm-exporter
  • VictoriaMetrics operator for the metrics path
  • Go 1.25+ to build from source
# CRDs, operator, secrets, TLS bootstrap, starter config — one run
# pin the version: it selects the matching signed images and manifest
curl -sfL https://github.com/kubeneuron/kubeneuron/releases/latest/download/install.sh \
  | bash -s -- --version latest

# ✓ KubeNeuron is Ready. → prints your panel sign-in and API token
kubectl -n kube-neuron port-forward service/kubeneuron-controller 8080:8080

Documentation

Read before you
trust it with a node.

Everything is written down: how it installs, what each XID means, what happens when a certificate expires, and what the system refuses to do. No sales-deck gaps.

Product tour Screenshots and a recorded demo of the whole workflow, from detection to approval. Quickstart A dry-run installation you can drive end to end, no GPU required. Production install Operator, TLS bootstrap, tokens, notifications, and the burn-in procedure. Playbook authoring Typed actions, approval gates, escalation ladders, and what admission rejects. XID catalog Every classified fault code, its problem class, and the reasoning behind the response. Operations Pausing, backups and restore, PostgreSQL HA, token rotation, host tooling. Alert runbooks One runbook per shipped alert — what fired, why, and what to do about it. Design document Architecture, trust model, safety invariants, and explicit implementation boundaries.

Build the safe path first

Run a calmer GPU fleet.

Read the source, run the dry-run ladder, and tell us where the policy model breaks.

Explore KubeNeuron Join the Discord