Privacy-first
Local transcription is the default path - nothing is uploaded unless you say so.
Privacy-first speech to text
Sawoice supports up to 35 languages with local models and 98+ with Scribe in optional Cloud Boost, which runs only when you enable it. macOS is stable; Windows 0.1.6 is a public beta.
Local transcription is the default path - nothing is uploaded unless you say so.
Fast transcription with on-device processing.
Runs offline after model setup, with Cloud Boost available when you choose it.
Add your terms and phrases for personalized accuracy.
Use up to 35 languages locally, or 98+ with Scribe in Cloud Boost.
How it works
Sawoice is designed to be beautifully simple so you can focus on what matters.
Capture speech with crystal-clear audio using Sawoice.
Fn on macOS · F8 on Windows
Local AI models process audio first, with Cloud Boost available when enabled.
On-device by default
Copy the transcript, keep it in local history, or paste it into your active app.
Clipboard-ready text
Language coverage
Keep supported languages on-device with local models, or explicitly enable Cloud Boost when you need its broader language set.
Combined local and optional Cloud Boost Scribe coverage
Keep everyday dictation on your machine by default.
Built for real voice notes, meetings, and long-form dictation.
Market, write, and work across teams without switching tools.
System requirements
Sawoice runs local AI models on-device. Requirements vary by model - start with Lite for the lowest footprint, or use Swift for the broadest language coverage.
Minimum
Recommended
Windows 0.1.6 is a public beta with a self-signed installer. Windows may show Unknown Publisher / Microsoft Defender SmartScreen warnings until trusted signing is added.
Minimum
Recommended
Total system memory needed with the model loaded and recording active.
| Model | macOS RAMUnified memory | Windows RAMSystem memory |
|---|---|---|
Lite Smallest local English model | 8 GB | 8 GB |
Core Balanced local English model | 8 GB | 8 GB |
Pro Higher-accuracy local English model | 16 GB | 12 GB |
Swift Fast local multilingual model | 16 GB | 8 GB |
Stream Live local multilingual model | 16 GB | macOS only |
Swift on Windows can use DirectML with a compatible GPU and at least 3 GB VRAM, but it falls back to CPU when that acceleration is unavailable. Built-in Apple Speech does not require a downloadable model.
Simple, transparent pricing
Start with private local dictation in up to 35 supported languages, then enable Cloud Boost Scribe when you need coverage across 98+ languages.

Lifetime local-only access.
Plan limits
Features

Local privacy plus Cloud Boost for daily work.
Plan limits
Features
Annual Cloud Boost access for active users.
Plan limits
Everything in Pro plus...

Fair-use Cloud Boost for heavy daily dictation.
Plan limits
Everything in Pro plus...
Annual fair-use Cloud Boost for heavy users.
Plan limits
Everything in Max plus...