Why run AI models locally?

Control over the model, the data path, and the tools around it. I run local models when they fit the work, and I keep my memory and working environment portable so one provider doesn't control the whole arrangement.

Local inference can keep data on infrastructure you control. Privacy still depends on the full setup: connectors, logging, network calls, and who can access the machine. It isn't free to run. Hardware, power, maintenance, and operator time all count. A hosted model can be the more sensible choice for a workload.

I care about a representative test, the complete cost, and a way to change direction. My Field Note on vetting local AI explains why the exact release, runtime, provenance, and license all need attention.