No. “On-device dictation” describes where speech becomes text. It does not, by itself, describe how a speech model is downloaded, how a license is checked, whether an app sends usage events, what an optional AI editor receives, or whether the document you paste into syncs online.
This distinction matters when a firm asks for offline recognition or an air-gapped workflow. Those are different requirements. In SuprFlow 1.0.0, WhisperKit recognizes speech on the Mac after its model has been downloaded. The default cleanup pass also runs locally. The same app includes account/license requests and release telemetry; choosing Cloud API formatting sends transcript text to a selected endpoint. Its Ollama formatter can be local or remote, depending on the configured address. Confirm this for the signed release you install with a network test using synthetic text.
Four questions to ask separately
| Question | What SuprFlow 1.0.0 does | What to verify |
|---|---|---|
| Does microphone audio go to a speech server? | The ordinary recognition path uses a downloaded WhisperKit model on the Mac. | Trace the installed signed build while dictating a harmless sentence. |
| Does the text go to an AI server? | Only when a network formatter is selected: Cloud API sends the pre-cleaned transcript and instructions; Ollama sends the same kind of text to its configured endpoint. | Check the provider, model, endpoint, and actual request path. |
| Does the app make other internet requests? | The app includes model download, account/license checks, release telemetry, and update-related behavior. | Capture the domains and purposes during installation, launch, dictation, and update. |
| Does the final note stay on the Mac? | SuprFlow pastes into the selected destination; that app governs its own sync and retention. | Check the account and storage location of the destination document. |
A downloaded speech model can support recognition without a connection while a license check or telemetry request still needs one. Likewise, an app can make a network request without uploading the audio or transcript. A domain list alone does not prove what a request body contains; confirming that takes payload inspection or an instrumented test build.
Choose the requirement your policy actually means
If the rule is “do not send client speech to a cloud recognizer,” test the microphone-to-text path. If it is “do not send client words to any remote processor,” also turn off hosted formatting, inspect any Ollama endpoint, and evaluate telemetry error fields. If it is “the application must operate for weeks with no network,” test installation, model availability, license grace, startup, and updates across that duration. If it is “no network packets at all,” a local speech model alone cannot satisfy that requirement.
For a practical pilot, use a fictional sentence that includes a made-up client name, amount, and negation. Record the signed app version, model, formatter setting, endpoint, destination app, and test time. Run once after a clean install and once after the model is present. Check both the inserted text and network behavior. Repeat after updating the app.
Questions people ask
Can SuprFlow transcribe without Wi-Fi?
The recognition and default cleanup paths are local once the model is present. That does not establish how long the installed signed app remains usable without account or license contact; test the offline license behavior for the exact release.
Does a local speech model make Cloud API formatting local?
No. Recognition and formatting are separate steps. Selecting Cloud API sends recognized text and instructions to the configured provider. The normal formatting request does not send the recording file in SuprFlow 1.0.0.
Is Apple's built-in Dictation always on-device?
Apple says to check the text below Dictation in Mac Keyboard settings to learn whether general voice input and transcripts are processed on that device. Availability and behavior vary by configuration. Check the Mac you are using rather than assuming all dictation products share one path.
Related guides
- Trace where SuprFlow dictation goes
- Does cloud formatting send audio or text?
- What usage telemetry can reveal
- Choose local or cloud formatting
- Test a signed release's network behavior