01
People with complex speech needs
ALS, aphasia, ICU, post-stroke, apraxia — anyone who can still blink or move their head.
App
Built for the way intention becomes speech.
From a passing idea to a meaningful exchange.
Sense → choose → speak
Live on iPhone and Apple Watch.
Full Grok conversations with Apple Health and live metrics as context. If speaking is hard, a sensing device or the front camera turns a blink, nod, or head shake into a spoken reply in your voice.
Join early access
Who it’s for
Built for decision, not performance.
Quiet software for hard moments — and for everyday Grok on a wrist.
02
Caregivers and clinicians
The phone hears the room, shows reply options, and speaks the chosen line out loud.
03
Anyone who wants Grok close
One tap, live voice, health and location in context. Quiet enough for a plane.
Two ways to talk
Talk to Grok. Talk to people.
Same product loop. Different rooms.
Talk to Grok
AssistantAsk about your health, search the web, check X. One tap starts a live voice session. Grok answers out loud. Tap to talk or hold to talk. Works on iPhone and Apple Watch.
Talk to people
Mind + cloned voiceBuilt for moments when speech is hard. The phone hears the other person, shows reply chips, and a cloned voice speaks the line you pick — so you can talk to the room.
Sensing deviceHead-worn sensing over Bluetooth. Connect in the app — no third-party pairing required.
Face cameraOn-device ARKit. No sensing device needed. Preview can hide; tracking stays on.
What ships today
Live voice. Real sensing. Soft Health UI.
Invite-only TestFlight. No public App Store listing yet.
Live Grok Voice (xAI realtime) on iPhone and Apple Watch
Tap to talk or hold to talk
Talk to people with reply chips + cloned TTS
Sensing device blink / nod / shake
Front-camera blink / nod / shake (ARKit, on-device)
Permissioned Apple Health, location, calendar, reminders
How it works
Sense → choose → speak.
Watch the demo, then the three steps. A sensing device is optional — gesture math stays on the phone.
Grokathon demo — blink, nod, or shake to pick a reply. Cloned voice speaks it.
01
Hear
Grok listens to you, or to the person in front of you.
02
Choose
Tap a chip, or blink / nod / shake with a sensing device or the camera.
03
Speak
Grok answers, or your cloned voice says the line to the room.
All gesture math stays on the phone. A sensing device is optional.
Stories
Why people are asking for this.
Families, caregivers, and clinicians — early voices from the waitlist.
I'm a parent building with the Grok stack because my son has childhood apraxia of speech. Standard AAC helps, but it's still slow and high-friction for real conversation. I got deep into Grok, Grok CLI, GrokTerm, and OpenClaw (Grok-powered agent) specifically to prototype an AI-assisted AAC path — including Grok Build for the boards/integration side. Dan's ThinkVoice direction matters: low-motor intent (face/head sensing) plus fast Grok Voice is exactly the kind of interface kids like mine need. If sensing + voice can cut the effort of “getting words out,” that's life-changing for families dealing with apraxia and complex communication needs.
I need this for my father who lost speech due to a stroke!!
In the ICU we spend so much energy trying to guess what someone is asking for. A phone that hears the room, offers reply chips, and speaks the choice out loud — with a blink or nod — would change how we care for patients who still know exactly what they want to say.
About
Wrist, room, and privacy.
Apple Watch
Same talk loop on your wrist.
Status, talk / stop, crown volume, short transcript. The phone is the hub for sensing device and camera. Sign-in syncs from iPhone.
Privacy
Permissions only when needed.
Microphone, camera, Bluetooth, location, and Health are requested only when a feature needs them. Face tracking runs on-device with ARKit. A sensing device never requires a third-party account. xAI auth stays in the Keychain. ThinkVoice does not invent faces of real people.
FAQ
Short answers.
Is this reading neural impulses or thoughts?
ThinkVoice works from headset sensor streams—mainly EEG, motion (accelerometers), and heart-rate data—rather than decoded thoughts. The same approach can extend to any device that can send a comparable set of signals.
How are responses generated? Is it reading my mind?
Replies aren’t pulled from private thoughts. A model proposes a range of options along a positive-to-negative spectrum; you choose one with a gesture, tap, or other input. Over time, optional profiling could tune defaults toward preferences and likely answers (for example food, phrasing, or common replies)—still as selectable options, not mind-reading.
Do I need a sensing device?
No. Talk to Grok works with a tap. Talk to people can use the front camera instead of a sensing device.
Is this a medical device?
No. Assistive voice software. Not a diagnostic product.
How do I get it?
TestFlight invite. Request access via the form — we’ll send an invite when a seat opens.
Join early access.
ThinkVoice is invite-only. Fill out the form and we’ll send a TestFlight link when ready.
Join early access