I’m confused, OpenClaw seems to be some kind of agent communication hub, but if you’re calling out to OpenAI or Anthropic, why would the model not have the same safeguards? If it’s a local model, how powerful does the hardware need to be to get results like this?