The promise
NutsNews continues to use the existing Qwen AI service. Its public address, authentication, model, and request format are preserved while the new infrastructure is added.
What is checked
The pipeline checks the AI endpoint and makes a small real inference request. Success must come from local Qwen. A response from a fallback provider does not count.
During other updates
Updating the portal or wiki does not restart Qwen or its API. If a change affects AI connectivity, the release is stopped or rolled back until the worker connection works again.
Workload visibility
The Qwen page now shows aggregate requests, errors, in-flight work, model availability, and latency. Initial instrumentation needs a guarded API-wrapper restart after two idle checks; it does not restart Ollama or change the model. If demand stays busy, deployment is deferred. Later unchanged portal releases do not restart the wrapper.