If you use open claw with a corporate LLM brain, you can ask it to interact with a local model. So they would see the query but not the answer which could stay on your machine