Dictation · checked 2026-10-14

Can I vibecode Superwhisper?

YES· one-shottable

Push-to-talk dictation is a short loop: record from the microphone, run a Whisper model on the device, paste the text where the cursor is. A Mac menu bar app that does this, plus a cleanup pass through a language model, is an afternoon for an agent, and the open-source field (Handy, VoiceInk, FluidVoice) shows that it works. Plan an hour for macOS permissions: Accessibility and Microphone grants follow the app's code signature, so a rebuilt app can lose them. The phone and Windows apps, Superwhisper's cloud models and its screen-aware Super mode are not part of the copy.

$8.49/moprice

$102you'd save a year

one sittingbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyIntermediate · Some system setup or keys.

Time2 to 4 hours in one sitting; macOS permissions take the longest

Claude Code, Codex or CursorAny of them can run the prompt.
A Mac with Apple SiliconSuperwhisper says offline models run well only on Apple Silicon; Intel Macs do better with cloud models.
Xcode command line toolsSwift 6 builds the package from the terminal.
A language model endpointOptional: a hosted key or local Ollama for the cleanup step.

Before you buildSuperwhisper Free already gives unlimited local Whisper dictation, and Handy is a free open-source option. Build only if you want your own modes or code.

1. Install one agent: Claude Code, Codex, Cursor.
2. Make an empty folder and open it in the agent.
3. Paste the prompt below, press enter, and answer its questions as it builds.

The prompt

Build me a Mac dictation app to replace Superwhisper. Requirements:

- A macOS 14+ menu bar app in Swift 6 and SwiftUI, built with Swift Package Manager; `make app`
produces Dictate.app in ./build/ and signs it ad hoc with the fixed bundle id com.local.dictate.
- Hotkey: hold to talk with the `KeyboardShortcuts` package (default Right Option, rebindable in
settings). Recording starts on key down and stops on key up; Esc cancels. A small floating pill
shows "recording" and "transcribing".
- Audio: AVAudioEngine at 16 kHz mono, kept in memory. Nothing is written to disk unless
KEEP_AUDIO is switched on in settings, then to ~/Dictate/audio/ with the date in the name.
- Transcription: the `WhisperKit` package with a model picker (base, small, large-v3-turbo)
that downloads on first use into ~/Library/Application Support/Dictate/models. Language set
to auto or a fixed code from settings.
- Cleanup, off by default: send the raw text to an OpenAI-compatible endpoint (LLM_BASE_URL,
LLM_MODEL, LLM_API_KEY in ~/.config/dictate/.env; local Ollama works) with the prompt of the
active mode: Plain (no change), Message (short, casual), Email (greeting and sign-off) or Notes
(bullets). A mode can be bound to an app by bundle id.
- Insert: put the text on NSPasteboard, send Cmd+V with CGEvent, then restore the previous
clipboard after 300 ms. Without Accessibility permission, leave the text on the clipboard
and show a notice.
- Vocabulary: ~/Dictate/vocabulary.txt (names and terms, one per line) goes to Whisper as the
initial prompt, and ~/Dictate/replacements.txt holds "heard => written" pairs applied afterwards.
- History: the last 200 results in ~/Dictate/history.json, listed in the menu to copy a past one.
- No accounts, no telemetry, local-first. Network only for the model download and the optional
LLM endpoint.
- Out of scope: Windows, iOS, Android, meeting recording, speaker labels, reading the screen.
- README: granting Microphone and Accessibility in System Settings, why a rebuild with a
different signature loses Accessibility (remove and re-add the app), and a 30-second test.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.

Share on X

Max MiskowSponsoredAI agents, automations and MVPs, built for your company.Visit site ›

Then keep going

Next prompts, same folder.

The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Superwhisper.

Add live transcription to the dictation app in this folder. Requirements:

- While the hotkey is held, run WhisperKit on the audio so far every second and show the
  running text in the floating pill.
- On release, run one final pass over the whole recording and paste only that result.
- If a live pass takes longer than the interval, skip the next one instead of queueing.
- Add a settings switch to turn live text off on slower Macs.
- Update the README.

Add voice commands to the dictation app in this folder. Requirements:

- After transcription, translate these phrases at the end or middle of a sentence:
  "new line", "new paragraph", "period", "comma", "question mark", "scratch that" (delete the
  last dictation by selecting it with Shift+Left and pressing Delete).
- A commands.txt in ~/Dictate/ lets me add my own "heard => text" pairs.
- Commands only apply in the Plain mode unless a mode turns them on.
- Unit-test at least 15 sentences; update the README.

Add translation to the dictation app in this folder. Requirements:

- A second hotkey (default Right Option + Right Command) records as usual, then asks the
  configured LLM endpoint to translate the transcript into TARGET_LANGUAGE from settings.
- Paste only the translation; keep both texts in the history list.
- If no LLM endpoint is configured, show a notice instead of failing silently.
- Update the README.

Add file transcription to the dictation app in this folder. Requirements:

- Dropping an audio or video file on the menu bar icon, or a menu item "Transcribe file",
  extracts the audio with `ffmpeg` (check it exists at start-up) and runs WhisperKit.
- Write ~/Dictate/transcripts/<file name>.md with timestamps every 30 seconds.
- Show progress in the menu and a notification when done; run one file at a time.
- Update the README with the ffmpeg install step.

What you lose

The iPhone, Android and Windows apps that share one licence
Cloud models and the extra on-device models such as Parakeet
Super mode, which uses what is on screen to pick the style
Meeting recording, file transcription and speaker separation
Modes that switch on their own for each app, tuned and maintained for you

Why people still pay

People pay for the finish: modes that follow the app in front, context from the screen, dictation on every device and a team that adds new speech models as they appear. The free plan already runs local Whisper without a cap, so Pro buys the AI layer and the extra devices. Threads we read show a lifetime owner moving to the free FluidVoice and others staying for the mode system.

Moatexecution polish

Do the math

Superwhisper or your own copy?

Over

Your usage

Count my time3h ×$/h

Superwhisper, 1 yr$102

Your own copy, 1 yr$23running + build session

You keep$79pays for itself in the first month

Estimate · Local transcription costs nothing. Cleanup through a hosted model is a fraction of a cent per dictation, or free with a local model. The build is token usage on your agent plan.

Leaving Superwhisper

Take your data, then cancel.

1 · Take your data out of Superwhisper

1Your history, custom modes, vocabulary and text replacements sit in the app folder on your own disk; the location is set in Settings under Advanced settings.
2On a Mac the recordings are in ~/superwhisper/recordings (older installs used ~/Documents/superwhisper/recordings). On Windows they're in the recordings folder under %LOCALAPPDATA%\com.superwhisper.app.
3Copy the modes' prompts into a text file if you want to reuse them in another tool.

2 · Cancel Superwhisper

1Web purchase: sign in at superwhisper.com/billing and choose Cancel subscription.
2iPhone: App Store > your profile picture > Subscriptions > Superwhisper. Android: Play Store > profile > Payments & subscriptions > Subscriptions.
3Pro stays active to the end of the billing period, then the account moves to Free.

The refund window is 30 days from the first purchase, and renewals aren't refundable on their own: cancel before the renewal date.
Nothing is deleted when Pro ends. Custom modes, vocabulary and history stay on the device, but Free keeps only local Whisper models and two modes without AI processing.
A lifetime licence is a one-time payment, so there is nothing to cancel; moving from a subscription to Lifetime refunds the unused part of the subscription through support.

Checked 2026-10-14 against Superwhisper's docs: 1, 2, 3, 4, 5

The escape hatch

Don't feel like building it? These are already free.

HandyOpen sourceMIT-licensed, fully offline speech-to-text: press a shortcut, speak, and the words are pasted into any text field. Whisper and Parakeet V3 models, with GPU acceleration where available.★ 34k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗VoiceInkOpen sourceGPL-3.0 native macOS 15+ app with offline transcription, per-app modes and a personal dictionary. The source is open; the maker sells a licence for the ready-built app with updates and support.★ 6.8k · active Oct 2026 · one-click install
macOSVisit ↗OpenWhisprOpen sourceMIT-licensed dictation with local Whisper or Parakeet models or cloud providers, plus meeting transcription and notes.★ 9.4k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗

All 4 free alternatives to Superwhisper ›

The numbers

Superwhisper pricing

Plan	Monthly	Annual, per month	What you get
Free	$0	$0	Local Whisper models without a cap, vocabulary, text replacements, two modes with no AI processing. Nothing dictated leaves the device.
Pro	$8.49	$7.08	Cloud models, the Parakeet and Cohere Transcribe local models, file transcription, unlimited modes, AI modes, context awareness, speaker separation, system audio recording, your own AI API keys. One licence covers unlimited Macs and Windows PCs plus an iPhone and an Android phone.
Pro Lifetime	custom	—	Same Pro features for a single payment.
Enterprise	custom	—	SOC 2 Type II, central billing and sign-in, model access control, enterprise-hosted models, volume discounts.

Free tier: Local Whisper models only, two modes and no AI processing; the cloud models, extra local models and AI modes need Pro.
Billing: Monthly, yearly or lifetime on the web; App Store and Google Play purchases are billed and refunded by Apple and Google. There is a 30-day refund window from the first purchase, and renewals are not refundable on their own.
Hidden costs: None found: cloud models are included in Pro, and your own API keys are optional.
Source: superwhisper.com

Also on the list

Wispr Flow$15/mo · DictationYES

Granola$14/mo · Meeting notesYES

Descript$24/mo · Audio & videoKINDA

Questions

Superwhisper, briefly

Not answered here? Tell us and we'll add it.

Ask a question ›

Free covers unlimited dictation with local Whisper models, custom vocabulary, text replacements and up to two modes without AI processing.

Yes. An AI coding agent such as Claude Code, Codex or Cursor can build a usable personal version in one session with the prompt on this page, running on your own machine.

About $8.49 a month on the Pro plan, $101.88 a year (checked 2026-10-14).

The iPhone, Android and Windows apps that share one licence; Cloud models and the extra on-device models such as Parakeet; Super mode, which uses what is on screen to pick the style; Meeting recording, file transcription and speaker separation; Modes that switch on their own for each app, tuned and maintained for you. If any of those matter to you, keep paying.