Story

In brief

What disaggregated serving actually buys you, how to run it end to end in vLLM today with prefill/decode plus the new GPU-less frontend and the things we're still working on.