💭 I think parallel mcts could be as simple as subclassing the request model function to work with batching making mcts use asyincio