All demos
Hylaq · Quiet

Type without typing.
Speak without sound.

Hylaq Quiet is the cross-vendor neural-input layer of Hylaq. Whisper into your phone — or in a year, mouth the words to your wristband, or in five, think them to your implant — and Quiet turns the signal into text, commands, and messages.

Subvocal · live

Hold the button, whisper at the lowest volume your throat will produce. Whisper-v1 + GPT-4o-mini cleanup turn it into clean text in ~2 seconds.

Voice for the Voiceless · live

AAC for ALS, locked-in, post-stroke. Mouth the words or tap a phrase; the device speaks. Free forever. The moral license that earns us the rest of the neural layer.

One pipeline, every signal

Signal
→ Decoder (signal-specific: Whisper, EMG-net, EEG-net, …)
→ Refiner (signal-agnostic LLM with user context)
Intent (text, message, command)

Modern LLMs are good enough that even very noisy signals (whispers, EMG sub-articulations, low-bandwidth EEG) decode reliably once the refiner has the user's recent context. This is why subvocal is tractable in 2026 and wasn't in 2020.

Signal sources

Audio (whisper-mode)
Live

Mouth the words at whisper volume. Whisper + LLM cleanup. Works in any browser today.

Surface EMG
Soon

Forearm wristband (Meta CTRL-Labs, Pison) — silent finger micro-gestures decoded as text.

Consumer EEG
Soon

AirPods-class scalp electrodes — focus state, attention level, basic intent.

Invasive BCI
Future

Neuralink / Synchron / Precision implants — full-bandwidth motor + speech decode.

Privacy posture

Raw signals are never stored. Only the refined transcript is kept, and only in your in-memory session for 24 hours.
Apps built on Hylaq Quiet receive decoded intents, not raw signals. Permission scopes are per-intent, not per-device.

For developers

One endpoint, every signal kind:

POST /api/quiet/decode
Content-Type: multipart/form-data

audio:  <Blob>         // today: webm/ogg/wav
handle: @colt          // optional, for session context
hint:   "send message" // optional, biases the refiner

→ { ok, text, raw, edited, decodeLatencyMs, refineLatencyMs }