For Mac Studio
Your inference box has one job. Keep the rest of the network out of its way.
Ollama, LM Studio, mlx, EXO: the API is a plain TCP listener on your 10 GbE port. FlowSense sees every client, every model pull, and every stray download competing with them, and lets you shape the lot.
See the clients
Which machine is hitting the API, at what rate, with what latency. Live and by the hour.
Protect the API
Give the inference port priority and cap model pulls, Time Machine, and browser traffic under it.
10 GbE ready
Passive capture at line rate, enforcement in the kernel packet path. Thunderbolt RDMA clustering stays untouched.