Description

After running nemoclaw setup-spark or rebooting the system, the OpenClaw gateway inside the sandbox is not running. There is no mechanism to auto-start it. Users must manually:

1. SSH into the sandbox via openshell ssh-proxy
2. Run openclaw gateway run
3. Set up an SSH tunnel to forward port 18789 from sandbox to host
4. Repeat all of this every time the gateway crashes or the system reboots

This makes integration with external dashboards (e.g. Mission Control) fragile. The gateway drops frequently and there is no supervisor or keepalive.

Expected Behaviour

nemoclaw start should also start the OpenClaw gateway inside the sandbox and set up port forwarding, alongside the existing Telegram bridge and cloudflared services. nemoclaw status should show the gateway state. nemoclaw stop should stop it cleanly.

Current Behaviour

nemoclaw start only starts the Telegram bridge and cloudflared tunnel. The gateway is not managed.

Environment

• DGX Spark (GB10, aarch64)
• NemoClaw via curl -fsSL https://nvidia.com/nemoclaw.sh | sudo bash
• OpenShell 0.0.7, OpenClaw 2026.3.11

Proposed Fix

I have a PR ready at #310 that adds openclaw-gateway and gateway-forward as managed services in start-services.sh. Happy to adjust the approach if the team has a different preference.