Dictation · checked 2026-10-14

Can I vibecode Wispr Flow?

YES· one-shottable

The loop is short: hold a hotkey, record from the microphone, run a local Whisper model, tidy the text and paste at the cursor. Open-source apps such as Handy and OpenWhispr already do this offline, so an agent can finish a working single-platform version in one sitting. The fiddly parts are macOS permissions (microphone, accessibility, input monitoring) that stop sticking when the app moves, and the cleanup quality. You give up the phone apps, sync, the meeting Notetaker and per-app tone.

$15/moprice per seat

$180you'd save a year

one sittingbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyIntermediate · Some system setup or keys.

TimeAbout an hour; macOS permission prompts take much of it

Claude Code, Codex or CursorAny of them can run the prompt.
macOS 13 or newer, Python 3.12 and uvThe prompt targets macOS only.
Microphone, Accessibility and Input Monitoring permissionsmacOS asks for each when the app first records, listens for the hotkey and pastes.
An OpenAI-compatible API key or local Ollama (optional)Only for the cleanup pass, which is off by default.

Before you buildWispr Flow Free allows 2,000 dictated words a week on desktop. If that is enough, you don't need to pay or build.

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 hold-to-talk dictation tool to replace Wispr Flow. Requirements:

- Python 3.12 on macOS 13+: a menu-bar app (`rumps`) started with `uv run dictate`,
  showing idle, recording and transcribing states.
- Hotkey: hold Right Option to record and release to stop (`pynput`). A tap
  under 300 ms is ignored; Escape while recording cancels.
- Audio: `sounddevice` at 16 kHz mono into memory, capped at 90 seconds; trim
  silence at both ends with `webrtcvad`. Audio is never written to disk.
- Transcription: `faster-whisper` with the `small.en` model, int8, fully local.
  A settings file selects another model for other languages. The first run
  downloads the model (about 0.5 GB) and says so.
- Cleanup: remove "um", "uh" and repeated words with rules and keep the
  punctuation Whisper returns. An optional LLM pass, off by default, goes to an
  OpenAI-compatible URL from .env (a local Ollama URL works) and only fixes
  self-corrections such as "five, no, six" without rewording.
- Paste: put the text on the clipboard, send Cmd+V with `pynput`, then restore
  the previous clipboard after 300 ms.
- Dictionary: ~/.dictate/words.txt of names and jargon, passed to Whisper as the
  initial prompt, and a snippets file that expands spoken triggers such as
  "my email" into text.
- History: sqlite3 at ~/.dictate/history.db keeping the last 500 transcripts,
  viewable from a menu item, with a "never store" setting.
- Single user. No accounts, no telemetry. The only network calls are the model
  download and the optional LLM URL.
- Out of scope: Windows, Linux, phones, meeting transcription, team features,
  per-app tone.
- README: the three macOS permissions to grant (Microphone, Accessibility, Input
  Monitoring), a warning that they attach to the terminal or the built app so
  moving it means granting them again, and a budget of 15 minutes for the prompts.

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 Wispr Flow.

Add a command mode to the dictation tool in this folder. Requirements:

- Hold Right Option + Right Shift to record a spoken instruction instead of dictation.
- Copy the current selection (Cmd+C, then restore the clipboard), send it and the instruction to the configured LLM URL, and paste the reply over the selection.
- Show the instruction and the result in a menu item for undo, and refuse to run if no LLM URL is set.
- Never send text unless the hotkey was pressed.
- Update the README.

Add per-app styles to the dictation tool in this folder. Requirements:

- Detect the frontmost app's bundle ID with `pyobjc` (NSWorkspace) when recording starts.
- A styles.toml maps bundle IDs to a style (casual, neutral, formal) and an optional list of words to avoid.
- With the LLM pass on, add the style as an instruction; with it off, only apply the rule-based cleanup.
- Log the app and style used with each history entry.
- Update the README.

Port the dictation tool in this folder to Windows 11. Requirements:

- Keep the core in a platform-neutral module and put hotkey, paste and tray code behind a small interface with macOS and Windows implementations.
- Windows: hold Right Alt via `pynput`, tray icon with `pystray`, paste with Ctrl+V, config under %APPDATA%\dictate.
- Document that apps running as administrator ignore simulated keystrokes from a normal process.
- Add tests for the platform-neutral cleanup and dictionary code.
- Update the README.

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

- While recording, transcribe the last 5 seconds every 2 seconds and show the latest text in a small floating window (`rumps` or a tiny Tk window) without typing it anywhere.
- On release, run the full transcription as before and paste only that result.
- Skip the live preview when the CPU is busy: if one pass takes longer than 2 seconds, stop previewing for the rest of the recording.
- Update the README.

What you lose

The iPhone and Android apps, and syncing between devices
Tone that adapts to each app, such as chat versus email
The meeting Notetaker with named speakers and summaries
Shared dictionaries, snippets and usage analytics for a team
Enterprise paperwork: SOC 2 Type II, ISO 27001 and a HIPAA BAA

Why people still pay

It types into every text field on four platforms, cleans up speech as you talk and learns your names and jargon, so the output needs little editing. People who dictate all day pay for that polish and for the same behaviour on laptop and phone. Occasional users can stay on the free 2,000 words a week.

Moatexecution polish compliance & regulation

Do the math

Wispr Flow or your own copy?

Seats

Over

Your usage

Count my time3h ×$/h

Wispr Flow, 1 yr$180

Your own copy, 1 yr$26running + build session

You keep$155pays for itself in the first month

Estimate · Local transcription costs nothing. The optional cleanup pass is billed per token by whichever provider you point it at. The build is token usage on your agent plan.

Leaving Wispr Flow

Take your data, then cancel.

1 · Take your data out of Wispr Flow

1The help center describes no export for dictation history or account data. On Mac and Windows, dictation transcripts sit in the app's local history, so copy anything you need before leaving.
2Dictionary entries and snippets live in the app's settings; copy them into a text file, because they don't carry over to a new account.
3Notetaker transcripts are stored in the cloud under their own retention schedule, which keeps them indefinitely unless you or your organisation set a window. Save the ones you need.

2 · Cancel Wispr Flow

1Mac or Windows: open Settings > Plans and Billing and select Manage subscription, which opens Stripe's billing portal, then cancel there.
2iPhone: Settings > your name > Subscriptions > Wispr Flow > Cancel Subscription.
3Android: in the Play Store go to your profile > Payments & subscriptions > Subscriptions > Wispr Flow > Cancel subscription. Cancelling inside Flow doesn't work for Play plans.
4Team plans: only admins can cancel, through Manage Subscription on the admin portal's Billing page.

Cancel where you bought the plan: deleting your account or signing out does not cancel an App Store or Google Play subscription.
Paid access runs to the end of the period, then the account returns to Free. Nothing is deleted.
There are no automatic prorated refunds; ask support, and Apple handles App Store refunds.
Deleting local history doesn't necessarily remove copies already uploaded to Wispr's servers. Use Settings > Data and Privacy to stop storing history going forward.

Checked 2026-10-14 against Wispr Flow's docs: 1, 2, 3

The escape hatch

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

HandyOpen sourceMIT-licensed desktop dictation that runs Whisper fully offline: hold or toggle a shortcut, speak, and the text is pasted into the focused field.★ 34k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗OpenWhisprOpen sourceMIT-licensed dictation with local speech-to-text models or your own cloud keys, plus meeting transcription and notes.★ 9.4k · active Oct 2026 · one-click install
macOS, Windows, LinuxVisit ↗VoiceInkOpen sourceGPL-3.0 native macOS app (macOS 15 or newer) that transcribes your speech almost instantly.★ 6.8k · active Oct 2026 · one-click install
macOSVisit ↗

All 5 free alternatives to Wispr Flow ›

The numbers

Wispr Flow pricing

Plan	Monthly	Annual, per month	What you get
Free	$0/user	$0/user	2,000 words a week on desktop and 1,000 on mobile, 100+ languages, personal dictionary and snippets, limited Notetaker access.
Pro	$15	$12	Unlimited dictation, longer Notetaker history, shared dictionary and snippets, a shared team workspace, centralized billing, usage analytics and priority support.
Growth	$23	$18	Everything in Pro plus SAML SSO, org-wide HIPAA enforcement with a signed BAA, multiple domains, admin control over model training and unlimited Notetaker access.
Enterprise	custom	—	Everything in Growth plus SCIM, audit logs, MDM, domain capture, custom contract terms and dedicated support.

Free tier: 2,000 words a week on desktop and 1,000 on mobile, with limited Notetaker access.
Billing: Monthly or annual (about 20% off annual). Pro has no commitment and can be cancelled at any time; plans bought through the App Store or Google Play are managed there.
Hidden costs: None listed for individuals. Dictation plans bought on a phone are cancelled with Apple or Google, not inside the app.
Source: wisprflow.ai

Also on the list

Superwhisper$8.49/mo · DictationYES

Speechify$29/mo · ReadingYES

Granola$14/mo · Meeting notesYES

Questions

Wispr Flow, briefly

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

Ask a question ›

Free allows 2,000 dictated words a week on desktop and 1,000 on mobile.

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 $15 a month on the Pro plan per seat, $180 a year (checked 2026-10-14).

The iPhone and Android apps, and syncing between devices; Tone that adapts to each app, such as chat versus email; The meeting Notetaker with named speakers and summaries; Shared dictionaries, snippets and usage analytics for a team; Enterprise paperwork: SOC 2 Type II, ISO 27001 and a HIPAA BAA. If any of those matter to you, keep paying.