The problem
Ask one model a question and you get one line of reasoning, often agreeing with however the question was framed.
What I built
A platform where Claude, Gemini, and OpenAI models act as advisors with different perspectives. They debate a question in a chat UI and converge on a shared answer. Conversation memory runs through the providers’ stateful APIs or a custom memory layer.
What changed
Answers get better when models challenge each other instead of agreeing by default.
What I learned
Independent agents are useful. Connected agents are the next step, so I’m building ways for them to share context and hand off work.