Yeah sadly, I'm pretty sure that api won't work on the castle environment , there's no way to do that rn from what I know What you can do is, making it from scratch by using large biases and weights