It's almost inevitable that it's going to happen, in a few months you won't be able to take a leak without an AI telling you to wash your hands. (Exaggeration)

Anyhow, technically doing this on the local machine is the way to go, checkout the 7b llama models that can be quantized down to 4bit. Something like that trained for games, run it on the background on a single thread.

It would not be super fast but it could chug away in the background analysing the game state and making decisions.