[kwindla]

Yeah, seems to be a drop-in replacement for the existing inference APIs. But I haven't found any docs yet for streaming audio/video input.