From reading their website, each bot gets its own neural network and its own reward function, so in that sense they are 5 separate agents. When training starts

From reading their website, each bot gets its own neural network and its own reward function, so in that sense they are 5 separate agents. When training starts out, the bots solely focus on their own reward functions. This is so they can learn very basic things like how to move around, how to attack, and so on. Over time, the combined team reward function gradually gets weighted more and more heavily, so that teamwork is encouraged.