Privacy is not a feature. It's the whole product.
Your voice and your vault never leave your Mac. When you use Pro rephrasing, only the transcript text is sent through a three-provider cascade configured to not retain your prompts. Here's exactly what stays local, what goes to the cloud, and what is never sent.
TL;DR
- Voice audio is transcribed on-device and discarded. It never touches the internet.
- Your vault (files, folders, tags, links) never leaves your Mac.
- Rephrasing (Pro, optional) sends only the transcript text to a Cerebras AI, Fireworks AI, Groq cascade, all configured to not retain your prompts.
- Both providers are contractually prohibited from training on your data.
- Telemetry is anonymous, opt-out, and every event is documented in full.
Who collects your data
The data controller for ThoughtMic is S3V Labs LLC, a Texas limited liability company (the company that publishes and operates ThoughtMic). You can contact us at hello@thoughtmic.com. ThoughtMic stores all dictation audio and transcripts on your Mac locally; the data we receive at all is limited to the items described below.
What stays, what's sent, what's never sent
Stays on your Mac
- Your voice recordings (transcribed locally by Whisper, then discarded)
- Your vault contents, file names, tags, and links
- Your dictation history
Sent to the cloud (only when you use rephrasing)
- The transcript text of your dictation
- The rephrasing instruction (e.g., "clean up grammar")
Never gets sent
- Your voice recordings
- Your vault content, files, or folder structure
- Anything identifying you beyond your license key
Rephrasing uses a three-provider cascade: Cerebras AI as the primary provider, Fireworks AI as the fallback, and Groq as the final fallback if both are unavailable. Cerebras's privacy policy states they do not retain inputs and outputs associated with their inference Services. Fireworks does not log prompts or generations on open-weights models by default. Groq has Zero Data Retention enabled on our organization. All three providers are contractually prohibited from training on your data. All three process data in memory and discard it after the response is returned.
How to verify this yourself
You don't have to take our word for it. ThoughtMic is designed to be inspectable:
Run Little Snitch. With rephrasing disabled, the app should make zero outbound network connections during dictation. With rephrasing on, you'll see one call to thoughtmic-worker.thoughtmic.workers.dev per rephrase, carrying transcript text only.
The telemetry and crash-reporting SDKs we use are open source: TelemetryDeck SwiftSDK and Sentry Cocoa SDK.
Your controls
In Preferences → Privacy you can:
· Turn off cloud rephrasing entirely (app works 100% offline)
· Turn off anonymous usage metrics
· Turn off crash reports (already off by default)
· Mark any dictation as Private; it skips telemetry and vault writes