Is Superwhisper safe? Local models, privacy and what it sends

Superwhisper is a commercial dictation app from SuperUltra, Inc., and it can run entirely on your Mac: pick a local voice model and either a local language model or no AI processing, and its docs say "no audio or text ever leaves your machine". It can also send audio or text to cloud models, chosen per mode. So Superwhisper is local when you set it up that way, and on the Free plan, which uses local Whisper models. It also lists a SOC 2 Type II report and offers a DPA and a BAA.

Publisher note: SuprFlow publishes this guide and is one of the products discussed. Competitor details come from each company's own pages, checked on 1 October 2026. Prices and features change; confirm before you buy.

This page reports what Superwhisper's own pages say. It is not a security audit, and we have not tested Superwhisper's controls or network traffic.

What do Superwhisper's own pages say?

As of 1 October 2026.

The Trust Center also names subprocessors: Anthropic, Cloudflare, AWS and Argmax (each listed as US) and OpenAI (no location given).

Does Superwhisper run locally and work offline?

Yes, with local models. Superwhisper's introduction says "Local voice models transcribe without an internet connection" and that offline AI formatting needs the local S1-mini model in the Message or Mail mode. Other AI processing needs the internet.

Which model Superwhisper uses depends on the mode. Its voice model page lists:

• Local, free: OpenAI's Whisper models, from Tiny to Large v3 Turbo, run through whisper.cpp.
• Local, Pro: Nvidia Parakeet models (through Argmax's WhisperKit SDK on macOS) and Cohere Transcribe (through MLX on Apple Silicon).
• Cloud, Pro: Superwhisper's S1-Voice, Deepgram's Nova models and ElevenLabs Scribe.

Two cautions. The Free plan says "Nothing you dictate leaves your device", but new users first get 3,000 words of Pro features, so check each mode's model during that period. And the sensitive data guide notes that dictating into a consumer service, such as the ChatGPT app, puts the text under that service's terms, whatever Superwhisper's own agreements say.

What do Superwhisper's pages not say?

• The privacy policy predates the cloud features. It is dated 19 June 2024, describes an app "that transcribes all audio data locally on your device", and says "We do not collect any usage data". It does not describe the cloud voice and language models, Filesync, or the Enterprise usage analytics that the current docs describe. Ask Superwhisper which document governs your use.
• Where cloud requests are processed. The pages we read do not state a processing region for cloud models, beyond the subprocessor locations listed in the Trust Center.
• What context awareness sends, in practice. The context docs say it can capture selected text, text from the active window and clipboard content, and that "unnecessary context sends extra information to the AI". Super Mode has all three on by default. The pages do not list exactly what reaches a cloud model in each mode.
• What the SOC 2 report covers. The report is behind a request, so its scope is not public.

How can you check for yourself?

1. Download from the right place. Superwhisper's quickstart says to download the Mac app from superwhisper.com and drag it to Applications. How to check a signed Mac release shows how to confirm who signed the app you installed.
2. Review permissions. The quickstart lists two: Microphone and Accessibility (to paste into other apps). Both appear in System Settings → Privacy & Security.
3. Open each mode. Note the voice model and the language model. A name with "(Local)" or a Whisper model is on-device; S1-Voice, Nova and Scribe are cloud.
4. Turn off context types you do not need in each mode.
5. Set Keep recordings for under Settings → Configuration if you do not want audio kept forever.
6. Run a network test with fictional text. Turn off Wi-Fi, dictate "Call Dana Whitlow about invoice 4471 for 2,950 dollars on Friday" in each mode you plan to use, and confirm the text still arrives. A mode that fails offline is using a cloud model.

When is Superwhisper a reasonable choice, and when does something else fit better?

Superwhisper is a reasonable choice when you want one app that can be fully local for sensitive work and cloud-assisted for everything else, when you need Windows or phone apps, or when a reviewer wants a SOC 2 report and a BAA. Its Free plan already gives local Whisper dictation at no cost.

Something else may fit better if a rule says nothing may ever reach a cloud model and you would rather not rely on each user choosing the right mode. Superwhisper's Enterprise plan lets admins disable hosted cloud models; an individual licence relies on settings.

A local-speech app with fewer paths is one option there. SuprFlow, for example, runs speech recognition on the Mac with a downloaded WhisperKit model, and its Default formatting is local cleanup of punctuation, capitalization and numbers. Its optional Cloud API formatting sends transcript text, not audio, to a provider you configure. It still uses the network for licence checks, model downloads, updates and usage telemetry. It is proprietary, Mac-only (Apple Silicon, macOS 14 or later), costs $125 once with no free tier or trial, and has no SOC 2 report, BAA, file transcription or meeting recorder. If you need those, Superwhisper does more. Where does my dictation go? traces SuprFlow's data path.

Questions people ask

Is Superwhisper legit?

Yes. It is an actively documented product from SuperUltra, Inc., with a privacy policy, a refund policy, a Trust Center and self-serve DPA and BAA signing.

Is Superwhisper open source?

Its pages do not describe it as open source, and we found no link to the app's source code on them. Some of the local models it runs, such as OpenAI's Whisper through whisper.cpp, are public projects linked from its model page.

Does Superwhisper work offline?

Yes, with local voice models. AI formatting offline needs the local S1-mini language model in the Message or Mail mode. Cloud voice models and other AI processing need a connection.

What model does Superwhisper use?

It depends on the mode you pick: local Whisper models on Free; Parakeet and Cohere Transcribe locally on Pro; and S1-Voice, Deepgram Nova or ElevenLabs Scribe in the cloud on Pro.

Related guides

• Superwhisper pricing
• Is Superwhisper worth it?
• Superwhisper alternatives for Mac
• SuprFlow vs Superwhisper
• On-device dictation vs no internet
• What a SOC 2 report proves for a dictation buyer

Sources

• Superwhisper — Privacy Policy
• Superwhisper Docs — Sensitive Data Best Practices
• Superwhisper Docs — Compliance and Trust
• Superwhisper — Trust Center
• Superwhisper Docs — Voice models
• Superwhisper Docs — Introduction
• Superwhisper Docs — Quickstart
• Superwhisper Docs — History
• Superwhisper Docs — Account and Data Deletion
• Superwhisper Docs — Context awareness
• Superwhisper Docs — Plans and Pricing
• Superwhisper — official site