On-Device Transcription
On-device transcription means speech processing runs inside your browser on your own machine. This page explains how the local-first model works and when it can operate offline.
Updated: February 17, 2026
How on-device transcription behaves
- You upload media in the web app interface.
- Model assets initialize in browser runtime.
- Transcription executes locally on your device.
- You export transcript output without uploading raw media payload to app APIs.
This approach is useful for teams that prioritize privacy boundaries and local control.
What local processing gives you
Privacy scope: Media processing stays in the local browser environment.
Operational control: Performance depends on your own hardware profile.
Offline continuity: Available after initial model cache is present.
Auditability: You can inspect network requests in DevTools during runtime.
Network behavior overview
| Request type | Typical reason | Contains raw media? |
|---|---|---|
| Model downloads | First-run setup and cache warm-up | No |
| Static assets | Load app HTML/CSS/JS resources | No |
| Licensing/payment APIs | Plan activation and billing | No |
| Transcription media payload | Core speech-to-text processing flow | Processed locally |
Related pages
FAQ
Does on-device mean zero internet forever?
Not exactly. Initial model download and some app services still need network, but core transcription can run offline after cache is ready.
Can I verify what is sent over network?
Yes. Use browser DevTools Network tab while transcribing to inspect request destinations and payload types.
Is this suitable for privacy-sensitive interviews?
It can be, because media processing stays local. You should still evaluate your device and browser security baseline.