How privacy mode works
In normal mode, Mavio sends your audio to cloud servers for processing. In privacy mode, the pipeline runs entirely on your device:| Step | Normal mode | Privacy mode |
|---|---|---|
| Audio capture | Local | Local |
| Audio upload | Sent to Mavio servers | Stays on your device |
| Transcription | Cloud AI models | On-device AI model |
| Speaker diarization | Cloud processing | On-device processing |
| Summarization | Cloud AI models | On-device AI model |
| Storage | Mavio cloud | Local device (optional cloud sync of text only) |
Enabling privacy mode
Desktop app
- Click the Mavio icon in the menu bar (macOS) or system tray (Windows).
- Click Privacy mode to toggle it on.
- Or use the keyboard shortcut:
Cmd + Shift + L(macOS) /Ctrl + Shift + L(Windows).
Per-recording
You can enable privacy mode for individual recordings without changing the global setting:- Before starting a recording, click the dropdown arrow next to the Record button.
- Select Record with privacy mode.
Always-on
To make privacy mode the default:- Go to Settings > Recording > Privacy mode.
- Toggle Always use privacy mode to on.
First-time setup
Privacy mode requires downloading the on-device AI model:Download the model
Mavio will prompt you to download the on-device transcription model. The download is approximately 1.5 GB.
Wait for installation
The model is installed locally. This takes 2-5 minutes depending on your internet speed and disk performance.
The model is stored locally and updated periodically. Updates are downloaded in the background and applied the next time you restart the app.
Accuracy comparison
On-device processing uses smaller, optimized models compared to cloud processing. Here is how accuracy compares:| Metric | Cloud mode | Privacy mode |
|---|---|---|
| English transcription | 97-99% | 94-97% |
| Non-English transcription | 93-97% | 88-94% |
| Speaker identification | 95-98% | 85-92% |
| Summary quality | High | Good |
| Processing speed | 2-5 min after meeting | 5-10 min after meeting |
Supported languages in privacy mode
The on-device model supports the following languages:- Full support (95%+ accuracy): English, Spanish, French, German, Portuguese, Italian, Dutch, Japanese, Korean, Mandarin Chinese
- Good support (90-95%): Russian, Polish, Turkish, Arabic, Hindi, Swedish, Norwegian, Danish
- Basic support (85-90%): Other supported languages
System requirements
Privacy mode uses local compute resources for AI processing:| Requirement | Minimum | Recommended |
|---|---|---|
| CPU | 4 cores | 8+ cores or Apple Silicon |
| RAM | 8 GB | 16 GB |
| Disk space | 2 GB free | 4 GB free |
| GPU | Not required | Metal (macOS) or CUDA (Windows) accelerates processing |
What data is stored locally
In privacy mode, the following data stays on your device:- Original audio recording
- Full transcript with speaker labels and timestamps
- AI summary and action items
- Voice profile data for speaker identification
Optional cloud sync
You can optionally sync text results (transcript, summary, action items) to the Mavio cloud while keeping audio local:- Go to Settings > Recording > Privacy mode > Cloud sync.
- Toggle Sync text to cloud on.
When cloud sync is enabled, only the text output is uploaded. The audio file remains exclusively on your local machine. You can delete the cloud copy at any time without affecting the local version.
Availability
| Platform | Privacy mode available |
|---|---|
| Desktop app (macOS) | Yes |
| Desktop app (Windows) | Yes |
| Mobile app | Coming soon |
| Browser extension | No (requires local model) |
| Meeting bot | No (audio captured in cloud by design) |