This is a good point. One potential workaround for this is to implement something usenet style where a whole corpus would be downloaded to your device and then

This is a good point. One potential workaround for this is to implement something usenet style where a whole corpus would be downloaded to your device and then you'd just load all the data locally. Of course only a small fraction of content would be available this way which has it's own set of biases.