Operations

You installed it. Now keep it running.

Health Checks

ClawMachine exposes GET /health which returns 200 OK when the server is up. Use it for Kubernetes liveness and readiness probes (the Helm chart sets these up automatically).

Each bot has its own health endpoint:

The dashboard shows bot status based on the Helm release state. If a bot’s pod is crashing, the status will reflect it.

Monitoring

ClawMachine doesn’t ship its own metrics endpoint (yet). For now, monitor at the Kubernetes level:

# Pod status
kubectl get pods -n claw-machine

# Resource usage (requires metrics-server)
kubectl top pods -n claw-machine

# Events — the first place to look when something's wrong
kubectl get events -n claw-machine --sort-by=.lastTimestamp

For bot namespaces, same commands with the appropriate namespace.

Upgrading ClawMachine

clawmachine upgrade

This upgrades the control plane only. Your bots are independent Helm releases — they keep running untouched. If a new ClawMachine version ships updated bot charts, existing bots don’t auto-upgrade. You’d upgrade them individually through the dashboard or helm upgrade.

Manual Helm fallback:

helm repo update
helm upgrade clawmachine clawmachine/clawmachine -n claw-machine

Upgrading Bots

From the dashboard, navigate to a bot’s detail page and use the upgrade option. Or via Helm:

helm upgrade <bot-release-name> -n <namespace>

Since bot charts are embedded in the ClawMachine binary, upgrading a bot to a newer chart version requires upgrading ClawMachine first.

Backups

ClawMachine itself is stateless — it reads everything from the Kubernetes API. No database to back up. If you lose the cluster, reinstall ClawMachine and re-create your bots.

What you should back up:

• 1Password vault — already handled by 1Password
• Bot data — if bots use PVCs (persistent volumes), back those up with your cluster’s backup solution (Velero, etc.)
• Custom Helm values — keep your values.yaml files in version control

Scaling

The ClawMachine control plane is a single deployment. You can bump replicaCount in the Helm values if you need redundancy, but honestly — it’s a dashboard. One replica is fine unless you have an SRE team that won’t let you sleep otherwise.

Bots scale independently through their own Helm values.

Security Notes

• ClawMachine uses the pod’s service account to talk to the Kubernetes API. It needs permissions to manage Helm releases and (optionally) ESO CRDs.
• The middleware stack adds Content-Security-Policy, X-Frame-Options, X-Content-Type-Options, and body size limits.
• Connect tokens are stored as Kubernetes Secrets, never echoed back to the UI.
• Static file serving has directory listing disabled.

Uninstalling

Remove a bot

From the dashboard detail page, or:

helm uninstall <bot-name> -n <namespace>

Remove ClawMachine

helm uninstall clawmachine -n claw-machine

Bots survive this. They’re separate releases. Remove them first if you want a clean slate.

Remove everything

# Uninstall all bots first, then:
helm uninstall clawmachine -n claw-machine
kubectl delete namespace claw-machine