Currently, token usage in Swarms is not aggregated for all the subconversations between agents. A simple utility function can be made to do that.