Unable to Run Nemoclaw and Build OpenClaw Agent After System Restart

[kuldeeprana09]

I powered off my system and then restarted it, but I still can't run Nemoclaw normally. I was able to rebuild the Hermes agent successfully. However, I cannot build the OpenClaw agent. I studied and followed the troubleshooting instructions provided in the NVIDIA Nemoclaw documentation, but the issue still persists. I am sharing the screenshots for reference. How can I fix these issues?

[BulletsHo]

Since Hermes rebuilt fine right after the restart but OpenClaw won't build, the base layer (Docker, the OpenShell gateway) is very likely healthy post-reboot — this looks like an OpenClaw-build-specific issue rather than the general host-reboot recovery path. Here's a checklist of the most common causes for exactly this symptom, sourced from NVIDIA's official NemoClaw troubleshooting docs, roughly in order of likelihood:

1. A previous onboarding attempt left a stuck session

If OpenClaw failed to build once already (network blip, Docker not fully up yet right after reboot, etc.), NemoClaw records that in ~/.nemoclaw/onboard-session.json and won't silently retry — especially when the installer runs non-interactively.

cat ~/.nemoclaw/onboard-session.json 2>/dev/null

If it exists, resume it (only useful if the earlier failure was transient):

nemoclaw onboard --resume

or wipe it and start clean:

rm ~/.nemoclaw/onboard-session.json
nemoclaw onboard

2. Port 18789 is still held by a leftover process from before the reboot

A dashboard SSH forward or an old openclaw-gateway process can survive an unclean shutdown. Re-onboarding then fails preflight with something like Port 18789 is not available. Blocked by: ssh.

sudo lsof -i :18789
kill <PID>          # kill -9 if it won't die

Then retry, or just pick a different port outright:

nemoclaw onboard --control-ui-port 19000

3. The build is failing during OpenClaw plugin install (web search)

If web search (Brave/Tavily) is enabled for the OpenClaw agent, the image build has to reach the npm registry / ClawHub during openclaw plugins install. A VPN or corporate firewall that hasn't reconnected since the restart will block this and fail the whole build. If the error in your screenshots is around plugin/npm install, confirm by disabling search:

NEMOCLAW_WEB_SEARCH_PROVIDER=none nemoclaw onboard --resume

If that succeeds, the network path to the plugin registry is the real problem — fix that route/policy, then re-enable search.

4. Docker's containerd/overlayfs snapshotter (Docker Engine 26+)

If the error mentions overlayfs snapshotter cannot be enabled or fuse-overlayfs, that's a known Docker 26+ default-driver conflict with k3s inside the OpenShell cluster image. Current NemoClaw patches this automatically, so it usually only shows up on an older NemoClaw build, or if NEMOCLAW_DISABLE_OVERLAY_FIX=1 is set somewhere in your environment:

echo $NEMOCLAW_DISABLE_OVERLAY_FIX   # unset it if it prints 1
nemoclaw update --yes && nemoclaw onboard

5. Stale gateway state

Worth a quick check either way:

openshell gateway list
nemoclaw <your-openclaw-sandbox-name> status

If the sandbox is registered locally but missing from the gateway, or the state looks inconsistent, this usually clears it:

nemoclaw <name> rebuild --yes

---

If none of these match, the fastest way to nail it down is the exact text of the error rather than a screenshot — could you paste the terminal output of:

nemoclaw onboard          # or: nemoclaw <name> rebuild --yes
nemoclaw host probe
docker version

plus which OS/platform you're on, and whether you're running onboard fresh vs rebuild on an existing sandbox? That should be enough to pin the exact cause.

Source: NVIDIA's NemoClaw OpenClaw troubleshooting guide (sections "Previous onboarding session failed," "Port already in use," "Sandbox build fails during OpenClaw plugin install," and "Reconnect after a host reboot") — https://docs.nvidia.com/nemoclaw/user-guide/openclaw/reference/troubleshooting