I just wonder whether the compute power is going to be there. Self hosting of any kind of generative model, especially something as big as an LLM sophisticated enough to act as an Agent, is at the very limit of what e.g. an Orin NX can handle.