Workflow update · 2 August 2026
Local voice.
Tailored output.
I have replaced Wispr Flow on my Mac with OpenSuperWhisper because Wispr Flow became unreliable on my desktop. I still use Wispr Flow on my phone. The result is no longer one cross-device product, but two tools chosen for the place where each currently works best.
01 / What changed
The trial is over. I switched the desktop.
The earlier version of this article recommended running both apps before deciding. I did that. Wispr Flow's desktop reliability became the deciding issue, while OpenSuperWhisper proved configurable enough to take over my everyday Mac dictation.
OpenSuperWhisper
Free, open-source dictation with local speech recognition and optional cleanup backends.
Maintained Mac download ↗Superwhisper
The similarly named commercial app now has Mac, Windows, and iOS versions. It is not OpenSuperWhisper.
Commercial iOS app ↗Wispr Flow
The cloud service remains my phone dictation tool even though I no longer depend on its Mac app.
Wispr Flow downloads ↗02 / My desktop stack
Local hearing, remote editorial cleanup.
My setup separates speech recognition from rewriting. Whisper hears the audio locally. A remote OpenAI model then corrects and formats the transcript according to my general and per-app instructions.
Whisper large-v3-turbo
OpenSuperWhisper records on the Mac and transcribes with the local large-v3-turbo model. The audio does not need to be uploaded for recognition, and dictation continues without a transcription service.
OpenAI gpt-4.1-mini
Cleanup uses OpenSuperWhisper's OpenAI-compatible remote backend. The transcript, not the audio, is sent to OpenAI for punctuation, obvious recognition fixes, and formatting. A separate selected-text Quick Action sends only the text I choose to OpenAI for translation.
Context for each app
Slack preserves mentions and commands, Terminal converts spoken symbols and identifiers, ChatGPT preserves technical syntax, and WhatsApp keeps a natural conversational style.
The privacy boundary
My audio stays on the Mac. My cleaned transcript does not.
Calling this setup completely local would be misleading. Recognition is local, but I deliberately opted into remote language-model cleanup. I can switch cleanup to OpenSuperWhisper's built-in model later if I want the full text pipeline to stay on-device.
03 / Controls and languages
Karabiner makes the Fn key my voice button.
OpenSuperWhisper keeps its normal Control-Return recording shortcut. Karabiner-Elements translates a simpler Fn-key gesture into that shortcut, so the application itself does not need a special global key combination.
Karabiner-Elements
Double to start, single to stop
- Double-tap Fn sends Control-Return immediately and starts recording.
- Single-tap Fn waits through a 300 ms double-tap window, then sends Control-Return to stop.
- Holding Fn still behaves like the normal function key.
- Karabiner is set to ISO for my physical Apple keyboards. This is keyboard geometry, not the Polish language setting.
- OpenSuperWhisper exposes a toggle, not separate start and stop commands, so the gestures are a convention rather than state-aware commands.
Polish and English
Translate after reviewing the English
- OpenSuperWhisper always inserts the cleaned English first. There is no spoken translation command.
- I review the English, select the text I want, then press Control-G.
- For a whole edit field, the sequence is Command-A, then Control-G.
- A macOS Automator Quick Action replaces the selection with natural Polish, a separator, then the exact reviewed English.
- Only the selected text is sent to OpenAI for this translation.
- Polish Pro is also active as a macOS input source for ordinary keyboard typing.
I will arrive at 6.→Control-G
Przyjadę o 6.+I will arrive at 6.
04 / Moving from Wispr Flow
I copied behavior, not an archive.
There was no single import button. I reviewed the useful Wispr Flow settings and recreated the parts that improve future output. I left behind data that would only preserve the past.
Dictionary and replacements
I brought across 41 reviewed dictionary records, including 28 active replacements. These cover names, technical terms, casing, abbreviations, and reusable spoken snippets.
App styles and cleanup
Wispr-style preferences became four OpenSuperWhisper App Context profiles for Slack, Terminal, ChatGPT, and WhatsApp, plus one general English cleanup instruction and a selected-text Polish translation Quick Action.
Audio and transcript history
I did not copy old audio files or convert the full Wispr Flow transcript archive. They add storage and privacy exposure without improving the recognition or cleanup of my next dictation.
05 / Mobile
My phone is still the gap in the open workflow.
OpenSuperWhisper remains a Mac application and has no native iPhone or Android companion. I therefore still use Wispr Flow on my phone. Commercial Superwhisper now offers an iOS app, so it is a real option to test, but it is a separate product rather than a mobile edition of the open-source application configured on my Mac.
MacWindowsiPhoneAndroid
OpenSuperWhisper
Wispr Flow
Fn gesture→OpenSuperWhisper→any Mac app
Local recognition, configurable cleanup, and per-app formatting.
phone keyboard→Wispr Flow→mobile apps
The familiar mobile workflow remains until I have tested a better replacement.
06 / Current comparison
The practical split in one table
CriterionOpenSuperWhisperWispr FlowCurrent choice
Current role
My desktop dictation system
My mobile dictation system
A deliberate splitReliability for me
Stable enough to replace Wispr Flow on my Mac
Became unreliable on my desktop, but still useful on my phone
OpenSuperWhisper on MacRecognition
Whisper large-v3-turbo running locally
Wispr's managed cloud pipeline
Control versus convenienceCleanup
Remote OpenAI cleanup with gpt-4.1-mini
Automatic cleanup managed by Wispr
Both produce polished textControl
Karabiner Fn gestures, plus a Control-G selected-text translation
Native app or mobile keyboard controls
OpenSuperWhisper on MacPolish output
Spoken triggers for Polish-only or bilingual output
Used as the mobile voice keyboard
OpenSuperWhisper on MacPrivacy boundary
Audio stays local; cleaned transcript is sent to OpenAI
Audio is processed by Wispr's cloud service
OpenSuperWhisper keeps audio localPlatform
Mac only, macOS 14+, Apple Silicon or Intel
Mac, Windows, iPhone, and Android
Wispr Flow on mobile07 / Verdict
OpenSuperWhisper won my desktop. Mobile remains separate.
OpenSuperWhisper on Mac, Wispr Flow on phone
This split gives me reliable, inspectable desktop dictation without forcing a premature mobile migration. It is less uniform than one synced service, but it matches what is actually dependable today.
Get the maintained Mac app ↗Whether the mobile half needs replacing
If Wispr Flow becomes unreliable on my phone too, commercial Superwhisper for iOS is now a concrete alternative to test.
This is not one synced system
OpenSuperWhisper settings stay on the Mac. Wispr Flow remains a separate mobile product with its own settings and data.