Onyx Voice / Web and API
Privacy notice
Effective date: October 29, 2026
Pilot status: Onyx Voice is operating as an invited web pilot. This notice describes the current product data flow; it is not a conclusion that the service satisfies a particular privacy law or data-residency requirement.
What the service processes
- Account details such as email address, display name, authentication identities, settings, invitations, and subscription state.
- Text submitted for speech generation and audio submitted for transcription.
- Authorized voice-cloning reference audio, voice metadata, rights attestation, retention choice, and consent state.
- Generated audio and generation metadata when Recent Generations is used.
- Usage counters, API-key metadata, request identifiers, and operational error information needed to run and secure the service.
How data moves
The browser communicates with the Onyx Voice Worker/API over HTTPS. The Worker enforces account ownership and forwards inference requests to configured TTS/STT services using an internal authentication token. Account metadata is stored in the configured D1 database; voice references, optional consented training archives, and generated audio use the configured R2 storage. The active deployment may also use Stripe for hosted billing, Google for optional sign-in, and an email provider for contact messages.
Retention and deletion
Voice source references use a selected 30-, 90-, or 365-day retention period and can be deleted by the owning account. A separate training archive is created only when training consent is explicitly granted; withdrawing that consent removes the stored training references. Generated audio is removed from storage before its history record is cleared. Browser drafts, generated audio cached in IndexedDB, voice samples, and recent transcription text may remain on the device until the user clears them or the browser removes site data.
Rights and synthetic output
Voice cloning requires the tester to attest to a valid rights basis. The service is designed for authorized source material and does not decide legal ownership for the user. Generated voice output is synthetic and is identified as such by the API. Do not upload source audio unless you have the authority to use it.
Residency and providers
Onyx Voice is built in Canada and priced in CAD. Processing and storage locations depend on the active Cloudflare, Modal, Stripe, Google, email, and other provider configurations. Canadian data residency is not claimed for a deployment until the active regions, contracts, and transfer terms have been verified.
Security
New API keys are stored as non-reversible digests. Older legacy key rows may remain until rotated. Worker-to-inference requests require an internal token, and user-owned records are checked against the authenticated account. No payment card number is stored by Onyx Voice; hosted payment surfaces are provided by Stripe.
Contact
For privacy questions or a deletion request, contact contact@onyxforge.app. Include the account email and the specific voice or generation involved, but do not email raw source audio or proof documents unless the operator gives you a secure intake method.