I've been experimenting with running NanoClaw on Kubernetes instead of macOS/Docker Desktop, and wanted to see if there's interest in upstreaming some form of this.

The core idea: replace the local runtime (WhatsApp, Apple Containers, filesystem IPC) with K8s-native equivalents, so NanoClaw can be deployed as a workload on a cluster.

• WhatsApp → HTTP webhook channel (receives events from any source)
• Apple Containers → K8s Jobs (each agent invocation is an isolated Job)
• Filesystem IPC → gRPC bidirectional streaming between orchestrator and agent pods
• Agent identity (CLAUDE.md, skills, settings) delivered via ConfigMap, per-group overridable
• Credentials via K8s Secrets, mounted as env vars through envFrom

The orchestrator runs as a Deployment. When a message arrives on the webhook, it creates a Job. The agent pod connects back via gRPC, receives the prompt, runs Claude, streams results back. Everything is externally configurable through env vars and ConfigMaps.

This makes it possible to run fleets of agents on shared infrastructure with proper RBAC, network policies, and observability ... "

I don't really want to turn this into a general orchestrator or job runner, but this seems like the bare minimum you would have to do to get this running on k8s.