And this is important because even though you are running a model locally, it's still a proprietary model. You have no say in what it was trained on, how that training data is labeled, what the guardrails are, what biases it might have, none of that.