Henry
In daily useA phone assistant that actually knows you
Tap-to-talk assistant that reads your messages, resolves your contacts, stages real actions, and refuses to make things up.
The problem
Voice assistants either do nothing useful or confidently invent things. The gap between a demo assistant and one you trust with your actual messages, contacts, and calendar is enormous, and it is mostly about honesty.
What we built
Henry is one tap on the side of the phone. Speak, and it answers in a steady British voice, grounded in a real search or in your actual data. When it cannot confirm something, it says so instead of improvising. That single behaviour, enforced with a per-claim gate, is what makes it usable.
It reads and summarises your real texts, resolves the right contact before it acts, and stages anything with consequences - a message, a calendar entry - for your confirmation before it fires.
Henry runs on our own gateway with the same memory and tool substrate as the rest of the fleet, which is why it keeps getting smarter without being rebuilt.
Touch it
Tap the orb to see a real exchange.
Tap the orb. This is a replay of a real exchange - scripted here, live on the phone.
One gesture
Tap, speak, done. On-device speech recognition finalises when you stop talking. No wake words, no modes.
Grounded or silent
Every factual claim passes an anti-fabrication gate. Unconfirmable answers are declined, not invented.
Your data, actioned
Reads real SMS, resolves real contacts, stages messages and calendar actions, and fires only on your confirm.
A voice worth hearing
Replies render server-side to a natural voice. One voice, every turn, no robotic fallback stacking.