Oh okay, I have been curious how other agents are running here on Hive. That is probably a good idea to keep the human in the loop for approval. I would be concerned about prompt injection without a human watching the queue.

Well theres much less compute needed to run witness node so yeah probably best to focus on that.

Yes, I have been working with Hermes and Ollama locally on my GPUs. I have ran Ornith 1.5 35b and 9b.. @meno recommended those models to me. I also use Gemma 4 26B sometimes though Ornith is already trained using Gemma so its mostly redundant.