Multi-agent setup

[BradKML]

Got the idea from the following application, feels like socratic/dialectic reasoning can be used to enhance the conversation, especially when web-crawling is allowed ala Perplexica https://github.com/satellitecomponent/Neurite

There would need to be some kind of template that juggles between different agents, including referees in case things can go south, and an "agent creator" that can spawn sub-agents of different personas based on the topic.