Auto-detects LM Studio (:1234) or Ollama (:11434) on
host.docker.internal / localhost; skips silently when no server is
reachable so CI stays green. When a server is present it exercises the
full predict_useful_life path through the OpenAI-compatible adapter,
catching prompt / JSON-parsing regressions that mocked LLMs hide.
Tagged 'local_llm' so it can be selected explicitly when an LLM is
known-available.
Made-with: Cursor