[Tostino]

Embedding models are generally lightweight enough to run on CPU, can be done in the background while the user isn't using their device.