Load the model, then run a task. A small model in your browser classifies the request and routes it to one of the bundled macros — which then runs locally against the live repo. Watch how much of your session is served with no cloud, no key.