Isn't that easily accomplished by setting up a local deployment and then yanking the network cable? Anything that can quickly run a capable LLM is going to be a pretty beefy box though. More like LLM in an expensive space heater.