Adapted community discussion: Paraphrased and condensed from an anonymized public OpenClaw Discord support conversation from August 2026. Technical guidance was reviewed for OpenClaw 2026.9.3.
I upgraded my Mac and now the Gateway does not load reliably. Doctor reports many warnings, the dashboard refuses connections, and I cannot tell which finding is the startup blocker.
Start with version and service-path evidence:
openclaw --version
which -a openclaw
openclaw gateway status --deep
openclaw status --all
A loaded LaunchAgent only means launchd registered it. A failed connectivity probe means the process never stayed on the port. Check for an old service path, plugin drift, or the first fatal config line before reinstalling anything.
The dashboard eventually loads, but all configured OpenAI models fail. Then the Gateway starts freezing and cycling on and off, so I cannot finish the model repair.
The logs show clean shutdowns requested with source=update, not random crashes. A custom launchd upgrade helper is marked keepalive and has respawned repeatedly. Preserve its latest logs, then unload only that stale helper job with launchctl bootout. Wait and confirm the restart log stops growing before starting the managed Gateway once. This leaves the script, config, and backups intact.
The Gateway is stable now. The remaining problem is model routing: my configured OpenAI models keep selecting an API-key route even though this agent uses OAuth.
The model names and OAuth profile are healthy. Inspect only the authored OpenAI provider and per-agent model overrides. If the provider contains a manually authored Platform API mode or base URL, remove only those route overrides, preserve the OAuth profile, then run openclaw models status --json --agent <agent-id>. Clear the old conversation pin with /model default -s or start a new chat.
That did it. The Gateway stays up and the OAuth-backed OpenAI route is working again.