In one line: everything except voice dictation runs on your Mac. Voice sends audio to OpenAI under your own API key, and neither OpenAI nor TypeNext keeps it or trains on it.
On your Mac vs. off your Mac
| Feature | Where it runs | What's sent |
|---|---|---|
| Autocomplete | On device | Nothing |
| Snippets & emoji | On device | Nothing |
| Personal dictionary | On device | Nothing |
| Read-aloud (TTS) | On device | Nothing |
| Screen context | On device (opt-in) | Nothing |
| Stats & learned style | On device | Nothing |
| Voice dictation | Cloud (OpenAI, your key) | Audio you dictate |
| Model downloads | Cloud (Hugging Face) | One-time file download |
| Subscription & billing | Cloud (Stripe) | Name, email, payment token |
Voice dictation, in detail
When you hold the dictation key and speak, here's the full path your audio takes:
- Audio is captured locally and sent directly from your Mac to OpenAI's transcription API, authenticated with the OpenAI API key you entered in Settings.
- OpenAI returns the transcribed text, which TypeNext optionally cleans up (also via OpenAI, your key) and inserts at your cursor.
- The request does not pass through Pocket Labs servers. We never see, store, or process your audio or transcripts.
No training, no retention
- Pocket Labs does not train on your data. We don't use your writing, audio, or transcripts to train models, and we couldn't, because they don't reach us.
- OpenAI does not train on API data. Data submitted through OpenAI's API is not used to train OpenAI's models by default. Where your OpenAI account supports it, TypeNext requests zero data retention so audio and transcripts aren't stored after the request completes.
- You hold the key. Because voice runs on your own OpenAI account, you control and can audit that usage directly, and you can turn voice off entirely so nothing leaves your Mac at all.
Permissions, and why
- Accessibility, to see the text field you're in and insert text. macOS blocks it from reading secure fields like passwords.
- Microphone, only for voice dictation, requested the first time you use it.
- Screen Recording, only if you enable screen context for an app; used to read on-screen text locally for better suggestions.
What you control
- Turn voice off, and nothing about your writing ever leaves the device.
- Block any app, disable TypeNext entirely in password managers, terminals, or anywhere else.
- Screen context is opt-in, per app, and off until you choose otherwise.
- Delete your data, clear snippets, dictionary, stats and learned style in Settings, or remove the app's data folder; uninstalling removes all on-device data.
- Opt out of diagnostics, anonymous crash/usage reporting is optional and never includes your content.
Security
On-device data lives in your Mac's protected app storage. Everything sent over the network, to OpenAI, Hugging Face, and Stripe, travels over encrypted TLS connections. We never sell your data; our business is selling software, not information about you.
Questions
Want more detail, or want us to confirm something for your security review? Email privacy@typenext.app. For the formal version, see the Privacy Policy and Terms.