I kept asking ChatGPT whether my ideas were any good, and it kept telling me they were. That is not a bug — a single model is talking to one person and has every incentive to agree. But it meant the one thing I actually wanted from it, a reason not to do something, was the one thing I never got. So I built Parley. You put several models in a Room. They answer your question independently, without seeing each other. Then they read each other's answers and argue. What comes back is separated into what they all agreed on and where they genuinely split — and the split is usually the interesting part, because that is where the judgement call lives. A few things I learned building it: - Models really do disagree, far more than I expected, once they can see each other's reasoning. - Forcing consensus destroys the value. The early version summarised everything into one clean answer and it was useless. Now disagreement is preserved as disagreement. - A model reviewing its own answer just agrees with itself. Peer review only works across models. Seven providers work immediately on Parley's own credits, so there's no API key setup. It's free to try. There's also a Chrome extension if you want to ask the models about a page you're already reading. I'm the developer and I built this alone, in Japan. I'd genuinely like to hear where it falls down — especially if you ask it something and the models all agree when they shouldn't.

Parley

Several AI models debate your question, then you decide

Several AI models debate your question, then you decide