Voice typing where audio
physically can't leak.
Whisper AI Chrome extension that runs locally — no servers, your audio never leaves your device. Built for HIPAA-conscious work. Works in Gmail, EHRs, ChatGPT — any text field.
How it works: local dictation in three keystrokes.
Install once
Add SafeDictate from the Chrome Web Store. Whisper AI (~90MB) downloads once and caches locally.
Click any text field
Open Gmail, ChatGPT, your EHR, or any page with a text box. Click to focus.
Press Alt+Shift+V
The recording overlay appears at the bottom of the page. Speak normally.
Text appears instantly
Press Alt+Shift+V again. Text is injected where your cursor was. Audio is gone from memory.
Private by design: your audio never leaves.
Most voice tools send your audio to Google or a private cloud. Even "offline mode" often means the audio is processed on their server first.
SafeDictate is different at the architecture level: Whisper AI is downloaded and cached locally in your browser. When you dictate, inference happens in a Web Worker — a sandboxed thread that has no network access. There is no server to send audio to.
- No API calls during transcription — network is disconnected after first load
- No account required — nothing to link your voice to your identity
- Private by architecture, not by promise — no audio leaves your device, so there's no data flow to regulate
Your audio never leaves this device
Transcription happens in your browser, offline after first load. No servers. No data flow to regulate. Not even analytics on what you dictate.
Private Voice Typing: SafeDictate vs. Wispr vs. Voice In.
| SafeDictate | Voice In Plus | Wispr Flow | Dragon Medical | |
|---|---|---|---|---|
| Price | Free + $1/mo | ~$50 one-time | $15/mo | $79–99/mo |
| Audio stays local | ✅ Always | ❌ Google cloud | ❌ Private cloud | ❌ Cloud version |
| HIPAA-conscious by design | ✅ No PHI leaves device | ❌ | ❌ | ⚠️ Enterprise only |
| Windows support | ✅ Chrome | ✅ | ✅ | ✅ |
| Mac support | ✅ Chrome | ✅ | ✅ | ✅ |
| Works in Gmail / EHR | ✅ | ✅ | ✅ | ❌ |
| No account required | ✅ | ✅ | ❌ | ❌ |
| Zero install (no backend) | ✅ | ✅ | ✅ | ❌ |
* Voice In uses Google WebSpeech API — audio goes to Google's servers.
Wispr Flow has a 'Privacy Mode' for Enterprise only ($30+/mo).
Your voice stays on your device. Plans start at $1/month.
No cloud. No account. No lock-in. Start free and upgrade when you need more.
- 30 min free trial
- Whisper tiny (multilingual)
- Gmail, ChatGPT, Google Docs
- No account required
- Unlimited dictation
- Whisper base (more accurate)
- Priority support
- Early access to new features
- Unlimited dictation
- Whisper base (more accurate)
- Priority support
- Early access to new features
Honest answers.
Is this actually HIPAA compliant?
HIPAA compliance depends on whether PHI (Protected Health Information) leaves the covered entity's control. SafeDictate processes all audio locally in your browser and never transmits it, so no PHI passes through us and there's no data flow to regulate. Whether a BAA is required depends on your overall workflow — confirm with your compliance officer. We make no warranty of regulatory compliance.
What happens on first install? Does it work offline?
On first use, Whisper AI model (~90MB) is downloaded once from HuggingFace CDN and cached permanently in your browser's Cache API. After that, everything works fully offline. No subsequent network calls are made during transcription.
How accurate is it?
The free tier uses Whisper Tiny — a lightweight model optimized for speed (2–5s latency on CPU). It handles everyday dictation well for English and Russian. The Pro tier will include Whisper Base, which is noticeably more accurate for medical and legal terminology. Neither model includes a specialized medical vocabulary — if that's a hard requirement, consider Dragon Medical.
Which websites does it work on?
Any text field you can click: Gmail, ChatGPT, Google Docs, Google Sheets, standard <input> and <textarea> fields across the web. SimplePractice, Jane App, and Notion support is being tested. If a site doesn't work, the text is copied to your clipboard with a toast notification.
What's the hotkey? Can I change it?
The default is Alt+Shift+V. You can override it in chrome://extensions/shortcuts. Custom hotkey settings directly in the extension are coming in a future update.
What data do you collect?
Nothing you dictate, ever — we don't run a server that could receive your audio or transcriptions. We see basic, anonymous install/uninstall events via Chrome Web Store analytics (not linked to any content). For the Pro tier, the extension sends your subscription email to our licensing endpoint, which checks it against Paddle — nothing else is sent.