phonebrain

Why it is paused

phonebrain asked whether the assistant needs a server at all. If a small model runs on the handset, the private things stay on the handset, it works on a train with no signal, and it costs nothing per answer.

We built it and ran real local models through it. The answer, honestly, was no — not yet. What fits in a phone's memory was not good enough for what we would want to put our name on, and a mediocre private assistant loses to a good one people already trust.

So it is paused rather than abandoned: this is a problem that hardware and small-model research are actively solving, and the day the on-device model is good enough, the rest of this is already built.