With a hardware DSP? It's gonna have software in it, but doing this kind of processing in the upper most top level OS stack is certainly a choice.
It seems like a good choice. It’s computationally extremely light and you can update it much more easily with new features (they actually did this once - to let you change the beamforming mode in the menu bar)
It is also notoriously time sensitive however, and while likely the hardware can already ensure the synchronization between mics, processing in the OS itself necessarily means buffering for a significant period so you don't run the risk of draining the pipe in a non-realtime system.