ZeroClaw OpenShift deployment

Deploy a minimal ZeroClaw agent on OpenShift with an external LLM provider (Anthropic, OpenAI, or any OpenAI-compatible API).

Prerequisites

Copy the sample manifests to create your real ones:

bash

for f in deploy-k8s/*-sample.yaml; do cp "$f" "${f/-sample/}"; done

Edit secret.yaml and replace REPLACE_WITH_YOUR_API_KEY with your actual API key
Update the image field in deployment.yaml to point to your registry (e.g., ghcr.io/youruser/zeroclaw:latest)
Update the namespace in all files if you want a different name
Optionally edit configmap.yaml to change the provider or model
Apply all manifests:
bash
```
oc apply -f deploy-k8s/
```

The real .yaml files are gitignored so your secrets and customizations stay local.

Check that the pod is running and the route is accessible:

bash

oc -n zeroclaw get pods
oc -n zeroclaw get route zeroclaw

Test the health endpoint:

bash

ROUTE=$(oc -n zeroclaw get route zeroclaw -o jsonpath='{.spec.host}')
curl -sf "https://${ROUTE}/health"

Send a test message:

bash

curl -X POST "https://${ROUTE}/webhook" \
  -H "Content-Type: application/json" \
  -d '{"message": "hello, what model are you?"}'

Edit configmap.yaml to change runtime settings:

After editing, re-apply and restart the pod:

bash

oc apply -f deploy-k8s/configmap.yaml
oc -n zeroclaw rollout restart deployment zeroclaw

State is ephemeral. Both state and workspace volumes use emptyDir — agent memory and session history do not persist across pod restarts. For production, replace these with PersistentVolumeClaims.
Vanilla Kubernetes. The Route object is OpenShift-specific. On vanilla Kubernetes, replace route-sample.yaml with a Kubernetes Ingress targeting port 42617.

bash

oc delete namespace zeroclaw