local-inference-lab

Capture a short live vLLM profile from an already-running workload. Use when the user says "take a capture", "capture vLLM", "profile vLLM", or asks to grab a GPU+CPU vLLM trace from the local OpenAI server at 127.0.0.1:8000, automatically bounded to 4 scheduler/engine steps, then copy rank 0 to lukea@orion.local:~/Desktop without probing or health-checking the instance.