A panel of AI agents that has to agree before you get an answer
It started with a problem in legal work: nobody trusts a single model's answer, and they shouldn't. So a chair agent gathers the context around a question first. The stakes, the values in play, how confident the answer needs to be. Then it briefs a panel of sandboxed models from different providers. They draft. The chair merges. They vote, and one dissent sends it back for another round.
Any agent can raise its hand and ask the operator a question. It's slow on purpose.