How it answers
It opens with what it can actually do for you, and every conversation stays on the phone.
What it brings
On-device by default
Brin ships with the on-device summary switched on: your question is answered by the engines on your phone, nothing is sent anywhere, and no language model is involved. The Connected Coach is a separate setting you turn on yourself.
It talks about your data
Name any swept measure and it reports what the sweep found, today’s reading against your range, and what moved with it.
It holds a thread
“How do you know?” and “What else moved around then?” resolve against what was just said.
It cites, and links
Every answer with a subject carries a link to the investigation behind it. Claims are checkable, not asserted.
It knows how it was measured
Ask about your nervous system and it will tell you the readings were all taken in one hour of the day, if they were.
Coach memory
Goals and preferences you save, used to shape what it offers.
Weekly review
What the week established, and what it didn’t.
Two modes, and you choose
| Mode | What leaves your phone |
|---|---|
| On-device summary the default |
Nothing. The question is routed to an engine on the phone and the answer is that engine’s output. No language model is involved, which is why it answers plainly rather than conversationally. |
| Connected Coach off until you turn it on |
Your question and a bounded evidence packet — the conclusions the engines already reached, plus a list of what was missing. Not your raw readings, and not your event titles. A language model uses it to decide which analysis to run and to phrase the reply; it still cannot originate a finding. |
The screen tells you which one answered, every time: On-device summary or Connected Coach.
What it will not do
It will not name a cause no engine established, will not diagnose, and will not fill a silence with something plausible. Asked a question the data cannot answer, it says the data cannot answer it.