Free · Open Source · On-device v0.15

Talk.
It types.

The fastest way to write on your Mac. Private, on-device dictation & meeting notes. Talk in any language.

Filler, grammar and spelling tidied on-device. Optional AI rewrite for deeper polish.

curl -fsSL https://openwispr.vercel.app/install.sh | sh
Download the DMG MIT licensed · macOS 14+ · Apple Silicon · also as .zip
The honest math
≈150
wpm · you speak
vs
≈45
wpm · you type

Roughly three times faster, before you count the backspacing.

Works everywhere you type

Any app. Any text field.

+ anywhere you can type

What it does to your words

You ramble. It writes.

Spoken
"um so uh i think we should um move the launch to thursday and uh tell the whole team about it i guess"
→
Typed
"I think we should move the launch to Thursday and tell the whole team."

Fillers stripped, sentences capitalized, your vocabulary corrected. All on-device, before it pastes.

Features

Six things, done right.

  1. 01On-device Whisper

    Runs on the Apple Neural Engine via WhisperKit. Your voice never touches a server. Works fully offline, no account, no bill.

  2. 02Live preview pill

    A floating pill shows your words as you speak, so you see the transcription before it lands. Release the key and it pastes at the cursor.

  3. 03Learns your words

    A custom dictionary plus corrections it remembers. Product names, teammates, jargon: teach it once and it types them right.

  4. 04Snippets

    Say "add my email" and your address pastes itself. Same for your LinkedIn, your booking link, your sign-off. Phrases you set once, in the words you actually say them.

  5. 0513 languages

    Auto-detected or pinned. Speak Hindi and get Roman script, or have it translated to English. Code-switch mid-sentence, it keeps up.

  6. 06Stats & history

    Every dictation saved and searchable, with your words-per-day stats. Per-app disable and sound cues keep it polite where you want quiet.

01 Track 01

100% on-device Whisper

Runs on the Apple Neural Engine via WhisperKit. Your voice never touches a server. Works fully offline, no account, no bill.

Snippets

Say it once. Never type it again.

A phrase fires. A bare word never does — "send me an email tomorrow" stays exactly as you said it.

You say → it types
  • add my emailyou@example.com
  • add my linkedinlinkedin.com/in/yourname
  • add my calendarcal.com/you/30min
  • add my signatureyour sign-off, in full

Each snippet takes as many phrasings as you need, because "add my linkedin" also comes out "add my linked in". Set your own in the Snippets tab.

Meetings

The minutes write themselves.

OpenWispr captures your mic and the call's system audio, labels who said what, and saves the transcript automatically.

  • You / Others speaker labels, from mic vs system audio. Works with Zoom, Meet, Teams, anything that makes sound.
  • Autosaves every transcript as Markdown, locally.
  • Optional AI summary: local via Ollama, or bring your own key. Skip it entirely and nothing ever leaves the Mac.

Saved to ~/Documents/OpenWispr

standup — tue 10:02 REC
You

I can take the migration if someone reviews the schema today.

Others

I'll review it after lunch. Do we ship Thursday or Friday?

You

Thursday. The launch email is already drafted.

Others

Works for me. Let's lock it.

Who it's for

Three people, one key.

The writer

Drafts emails and docs at speaking speed, then edits instead of typing from scratch.

The engineer

Dictates commit messages, PR reviews, and prompts into Cursor without breaking flow.

Meetings all day

Lets transcripts and summaries pile up in a folder while actually listening on the call.

Compare

The honest table.

Normal people don't want an API key or a monthly bill. OpenWispr cleans your words on your own Mac, for free.

OpenWispr Wispr Flow Muesli
On-device✓✗ cloud✓
PriceFree$15/mo Pro*Free
Meetings✓✗✓
AI cleanupFree, on-device*CloudYour own API key
Cleanup, zero setup✓✗ paid cloud✗ needs API key
Open sourceMIT✗MIT
Works offline✓✗✓

*Cleanup runs on-device, built-in on macOS 26 with Apple Intelligence; on older Macs bring an API key or run Ollama (both optional). Wispr Flow's free tier caps at 2,000 words per week and processes audio in the cloud.

Install

Running in three steps.

01

Paste one line

Installs to /Applications and launches. No Gatekeeper dance — the script strips quarantine for you.

curl -fsSL https://openwispr.vercel.app/install.sh | sh
02

Walk the short setup

A seven-step wizard: mic + live level meter, Accessibility, speech model (~1.5 GB once), press-test your hotkey, then a practice dictation.

03

Hold the key. Speak.

Release — clean text lands at your cursor. OpenWispr lives in the menu bar, not the Dock. Rebind the trigger anytime in Settings.

Prefer the DMG? Download it, drag to Applications. On macOS 15+ Gatekeeper may block the first launch: open System Settings → Privacy & Security and click Open Anyway. Once.
FAQ

Fair questions.

I clicked the app and nothing happened?›
OpenWispr lives in your menu bar (microphone icon, top right) — there is no Dock icon. Clicking the app in Applications opens its home window; the menu-bar icon’s “Open OpenWispr” does the same.
Why does macOS warn me when I first open it?›
OpenWispr isn’t notarized through a paid Apple Developer account, so Gatekeeper flags it like any unsigned open-source app. The install one-liner handles this for you. For the DMG: System Settings → Privacy & Security → Open Anyway, one time only. The code is public on GitHub if you’d rather read it first.
How big is the download?›
The app itself is a few megabytes. On first setup the wizard downloads the Whisper model, roughly 1.5 GB, once. After that everything runs locally with no further downloads.
What does first-run setup look like?›
A short wizard: microphone with a live level meter, Accessibility for paste and the fn key, the speech model, a press-test of your trigger, and an optional practice dictation. Meeting capture is optional and can wait.
Does any audio or text leave my Mac?›
No. Transcription runs on-device via WhisperKit. There’s no account, no telemetry, no server. The only network calls are the one-time model download and, if you opt in, your own AI endpoint (local Ollama or a key you bring).
Do I need Ollama?›
No. Ollama is optional, only for AI meeting summaries. Dictation, filler removal, dictionary, snippets, and transcripts all work without it.
What permissions does it need, and why?›
Microphone to hear you. Accessibility to paste text at your cursor. For meeting transcription, screen & system audio recording so it can capture the other side of the call. Nothing else, and each is requested only when the feature needs it.
Ready when you are

Your voice. Your Mac. Nothing else.

Download the DMG Free forever · MIT · macOS 14+ · Apple Silicon