Dhavalu585/llm-council — reverse-engineered prompt

Build me a simple local web app that feels like ChatGPT, but instead of asking one AI, it sends my question to a small council of different models through OpenRouter.

I want to type one prompt and then see the process in three parts. First, each model gives its own answer and I can flip through them in a tab style view. Then each model reviews the other answers without knowing which model wrote them, and ranks them for accuracy and insight. Finally, one chosen chairman model combines everything into a single final answer for me.

Please make it easy to run locally, using an API key from a root .env file, and let me change which models are in the council and which one is the chairman. It should save conversations locally so I can come back to them later. If there is already starter code here, finish it and make the full flow work end to end. You can check current docs online if needed.