#12049 [DGX Spark][CLI&UX] tunnel stop reports Ollama cleanup failure on a vLLM-only installation

Description

On a fresh, working local-vLLM installation, nemoclaw tunnel stop reports an Ollama model cleanup failure and tells the user to restore an Ollama endpoint, even though Ollama was never installed or selected. The command returns exit code 0 while reporting incomplete cleanup.

The selected provider is vllm-local. Onboarding, inference, and an OpenClaw TUI conversation succeeded before this command. The subsequent full uninstall succeeded; this report concerns the misleading backend cleanup diagnostic.

Platform scope: Reproduced on DGX Spark only; other platforms not tested Regression: Unknown — earlier versions not tested OpenShell issue: No

Environment

Device:        DGX Spark (NVIDIA GB10)
OS:            Ubuntu 24.04.4 LTS
Architecture:  aarch64
Node.js:       v22.23.2
npm:           10.9.8
Docker:        29.6.2
OpenShell CLI: 0.0.116
NemoClaw:      v0.0.124
OpenClaw:      2026.7.1

Provider:      vllm-local
Model:         nvidia/NVIDIA-Nemotron-3-Nano-4B-FP8
Ollama:        Not installed or configured; no local Ollama listener
Public tunnel: Not configured

Steps to Reproduce

Start on a DGX Spark with working Docker access and no existing NemoClaw, OpenShell, Ollama, or Docker containers.

Install NemoClaw with managed local vLLM:

curl -fsSL https://www.nvidia.com/nemoclaw.sh -o nemoclaw.sh
env NEMOCLAW_INSTALL_TAG=v0.0.124 \
    NEMOCLAW_AGENT=openclaw \
    NEMOCLAW_PROVIDER=install-vllm \
    NEMOCLAW_NON_INTERACTIVE=1 \
    NEMOCLAW_YES=1 \
    NEMOCLAW_POLICY_MODE=suggested \
    NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=1 \
    NEMOCLAW_SINGLE_SESSION=1 \
    bash nemoclaw.sh

Let onboarding complete. Verify the sandbox is Ready and uses vllm-local:
```
nemoclaw my-assistant status
```
In this run the selected model was nvidia/NVIDIA-Nemotron-3-Nano-4B-FP8. An OpenClaw TUI conversation also returned a successful response.

Stop services and capture the command's exit code:

nemoclaw tunnel stop
printf 'TUNNEL_STOP_EXIT=%s\n' "$?"

Expected Result

Because this installation uses only vLLM and has no Ollama service, stopping services should skip inapplicable Ollama cleanup. It should not tell the user to restore an Ollama endpoint or report incomplete Ollama cleanup.

Actual Result

The command reports that the in-sandbox OpenClaw gateway stopped, then prints an Ollama cleanup failure, instructs the user to restore the unconfigured endpoint, and finishes with:

[services] Host services stopped; Ollama model cleanup remains incomplete.
TUNNEL_STOP_EXIT=0

Observed once after a fresh installation. A subsequent nemoclaw uninstall --yes returned exit code 0. Independent checks after that uninstall found no running Docker containers, no related listening ports, and no GPU workloads.

Logs

[services] Stopping in-sandbox OpenClaw gateway (sandbox: my-assistant)...
[services] OpenClaw gateway stopped inside sandbox.
[services] Ollama model cleanup failed at http://127.0.0.1:11434 (discovery-failed: No reachable local Ollama endpoint was found for cleanup). The saved local route was retained; restore access to http://127.0.0.1:11434, then retry this command.
[services] cloudflared was not running
[services] Host services stopped; Ollama model cleanup remains incomplete.
TUNNEL_STOP_EXIT=0