How HyperVoice Keeps Your Data Private
Privacy is not a feature we bolted on — it’s the foundation of how HyperVoice works. Your voice is personal. What you dictate at work, in medical notes, in private messages — none of that should pass through someone else’s server unless you explicitly choose it.
Here’s exactly how HyperVoice handles your data at every step.
Your Audio Never Leaves Your Device
When you press the hotkey and speak, HyperVoice captures audio from your microphone and holds it in memory. The AI speech recognition model — running entirely on your CPU or GPU — transcribes that audio into text. Once the transcription is complete, the audio is discarded. It is never written to disk, never uploaded, and never stored.
This isn’t a “we promise not to look” policy. The audio physically cannot leave your machine because the entire speech-to-text pipeline runs locally. There’s no network request involved in the transcription step.
On-Device AI Speech Recognition
HyperVoice uses state-of-the-art AI speech recognition models that run entirely on your device. The model files are downloaded once and stored locally on your machine. After that, transcription works completely offline — no internet connection required.
The models range from 75 MB (Tiny) to 3.1 GB (Large-v3). Larger models are more accurate but require more resources. All of them run locally, regardless of size.
GPU acceleration is powered by Vulkan, which works with NVIDIA, AMD, and Intel graphics cards. This keeps transcription fast without requiring any cloud computation.
What Happens with AI Post-Processing
HyperVoice includes optional AI post-processing modes that can clean up grammar, reformat text, or transform your dictation into structured output. These modes require an AI provider — and this is the one place where data leaves your device, only if you opt in.
Here’s what happens depending on the provider you choose:
No Provider (Default)
If you don’t configure a provider, post-processing is skipped entirely. Your raw transcription is pasted as-is. No data is sent anywhere.
Bring Your Own Key
When you use your own API key, HyperVoice sends the transcribed text (never the audio) directly to the provider’s API. The request goes straight from your machine to the provider — it does not pass through HyperVoice servers. Your API key is stored locally in your operating system’s secure keyring.
What the providers do with your data is governed by their own privacy policies. Most major AI providers offer API data policies that are separate from their consumer products — API data is typically not used for training.
HyperVoice Cloud (Pro Subscription)
With HyperVoice Cloud, the transcribed text is sent to our processing endpoint, routed to an AI provider, and the result is returned to your app. We do not store, log, or retain the content of your text. We log basic metadata (which model was used, token counts, timestamps) for service health monitoring, but your actual words are never persisted.
What We Do and Don’t Collect
Here’s a clear breakdown:
| Data | Collected? | Details |
|---|---|---|
| Audio recordings | Never | Processed in memory, immediately discarded |
| Transcription text | Never stored by us | Only sent to a provider if you opt in |
| API keys | Stored locally only | In your OS keyring, never transmitted to us |
| License key | Yes | For activation and validation via Lemon Squeezy |
| Email and name | Yes | From license purchase, for account management |
| Usage metadata | Optional | Feature usage, crash reports (no text content) |
| Website analytics | Yes | Google Analytics on hypervoice.app |
License Activation and Offline Use
HyperVoice requires a license key to activate. Your license key is validated once online via Lemon Squeezy, our billing partner, and then cached locally. After activation, the app works fully offline for up to 30 days before re-validation is needed. Your license key and instance ID are stored in your operating system’s secure keyring — they never leave your device beyond the initial validation request.
Open-Source Foundation
The core speech-to-text engine is built on open-source components:
- Open-source speech recognition models — State-of-the-art, community-audited speech-to-text
- High-performance C++ inference engine — Optimized for local execution
- Vulkan — Cross-vendor GPU acceleration
You don’t have to trust a black box. The transcription technology is open, auditable, and widely used.
Our Privacy Policy
For the full legal details, read our Privacy Policy. The short version: we collect the minimum data needed to operate the service, we don’t sell your data, and your audio never touches our servers.
If you have questions about how your data is handled, reach out at support@hypervoice.app.