Technically ,the nice thing about this is that they use an event based model. Your system is divided into event-manager threads which communicate with each other using events(and of course interrupts create events). When they are activated by an event , they do their stuff rapidly - and go to sleep. Only a single thread is allowed at a time. There are also event priorities to manage a case of multiple events happening at the same time, and a scheduler to manage that.

This enables both low power systems(since both threads and scheduler are always asleep unless there's a new event - unlike traditional rtos, where usually the scheduler needs to wake up periodically quit often to check which process's turn is now and give it to him).

And since threads are run to completion, you only need a single stack(instead of stack per task), which saves a lot of ram, which is critical in those kinds of systems, both in terms of price and in terms of sleep power(since even in sleep you need to power the ram and the smaller the ram the better) which usually is the biggest component in total power consumption.