I've built similar a couple of different ways. I'm currently focusing on running outputs through a local model to to enable tokenmaxing before sending onto a model I am paying with credits, not OAuth.