“I can’t see anything until I stop talking.”
Your words appear in the text box as you say them.
Murmur types word by word while you talk. When a sentence locks in, any word it misheard corrects itself in place. You catch mistakes mid-sentence.
Mac dictation that runs on your Mac
It types while you’re still talking.
Every word lands in your app 155 ms after you say it. Spot a mistake mid-sentence, not after. No cloud, no account, no word limits.
Priya
Is the login fix going out today?
Message #launch
Yes, it’s merged. Can someone review the pull request before lunch?
- 155 msfrom your voice to the screen, per word
- 3×faster live words than Apple’s engine
- 9.4%word errors, vs 18% for Apple’s transcriber
- 0bytes of audio leave your Mac
Why people give up on dictation. Fixed.
The complaints that come up again and again about dictation apps, and what Murmur does instead.
“It sends my voice to someone else’s server.”
Speech is recognised on your Mac.
No account, no upload, no audio stored anywhere. Turn Wi-Fi off and keep talking.
Wi-Fi off · still works
“The app went down and lost what I said.”
There’s no server to go down.
Nothing to time out, no audio chunks lost on a bad connection. It works on a plane, in a basement, on hotel Wi-Fi.
“It cut off my first word.”
Start talking the moment you press the key.
The mic stays ready and keeps the last 0.3 s before you press, so your first syllable is in the text.
0.3 s pre-roll
“It mangles names and tech words.”
It knows the words developers say.
versalVercelcloud codeClaude Codelead codeLeetCodeheadabadHyderabad
Add your own names in Settings.
“I still reach for the mouse to hit Send.”
It presses Return for you.
In Claude Code, ChatGPT and Slack your message sends when you let go. Say “Murmur, open Antigravity” or “next song” and it does that too.
⏎ sent
“It doesn’t get how I speak.”
Tuned for Indian English. Telugu too.
Murmur notices when you switch to Telugu and writes it in Roman letters, ready for WhatsApp.
“I hit the free plan’s word limit.”
No word limits. Ever.
Nothing runs on a server, so there’s no meter running. Dictate a 3,000-word essay on a Tuesday.
Same sentence. Three engines.
Each lane shows the words at the average delay we measured for that engine on the same recording.
You say
- +155 ms
- +462 ms
- +725 ms
Murmur shows each word 3× sooner than Apple’s live engine, and 4.7× sooner than its transcriber.
Measured with Murmur’s own benchmark (murmur-bench): one English recording streamed at real-time speed, each word’s delay checked against a time-aligned reference. Medians: Murmur 147 ms, Apple 455 and 742 ms. The lanes replay those averages.
This is the real caption bar.
It floats at the bottom of your screen. Grey words are still being heard; when you let go they lock in.
How Murmur compares
The big names in Mac dictation, from their own websites, checked 27 October 2026.
| Feature | Murmur | Wispr Flow | Superwhisper | VoiceInk | Apple Dictation |
|---|---|---|---|---|---|
| Where your speech becomes text | On your Mac | In the cloud1 | On your Mac or in the cloud, by model2 | On your Mac3 | On your Mac or Apple’s servers, by language4 |
| Types into your app while you talk | Word by word, on by default | Not stated | “Realtime” on supported models5 | Not stated | Yes |
| Word limit | None | 2,000 words a week on the free plan1 | Free plan; Pro adds larger models2 | None stated | None |
| Runs your Mac by voice: apps, music, tabs | Yes | Not stated | Not stated | Not stated | Separate feature (Voice Control) |
| Price | Early access | Pro $15 a month, or $12 billed yearly1 | Pro monthly, yearly or lifetime2 | From $25, one time3 | Built into macOS |
| Platforms | Mac | Mac, Windows, iOS, Android | Mac, Windows, iOS, Android | Mac, iOS | Mac, iPhone, iPad |
- Wispr Flow: “Transcription always occurs on the cloud” and its pricing page.
- Superwhisper: “Superwhisper works offline”, with cloud and local models; plans on its homepage.
- VoiceInk: “processes all voice transcription locally”; Solo licence $25.
- Apple: Use Dictation on Mac.
- Superwhisper: Realtime.
Say “Murmur” and it runs your Mac
Open apps and folders, play music, change the volume, switch tabs, lock the screen. Several at once. When it isn’t sure, it asks first.
- “Murmur, open Antigravity in my murmur folder”Opens it
- “Play Tum Hi Ho on YouTube”Starts playing
- “Open WhatsApp and close Safari”Does both
- “Open cursr”Open Cursor?
- “Next song”, “volume up”Media keys
Hands-free when you want
- Right ⌘ hold
- Talk
- Right ⌘ twice
- Hands-free
- Esc
- Cancel
- “new line”
- Line break
Questions
Something else? Email me.
Which Macs does it run on?
macOS 27 on Apple silicon. On first launch it asks for Microphone and Accessibility permission, so it can hear you and type into other apps.
Does it really work offline?
Yes. Dictation runs entirely on your Mac. Only voice commands that need the web, like playing something on YouTube, use the internet.
Which apps can it type into?
Any app with a text box: Slack, Notes, Mail, your browser, Claude Code, ChatGPT, your editor. Type-as-you-speak works in text boxes and chat or coding apps; elsewhere the text goes in when you let go.
How is it so fast?
Captions come from NVIDIA’s Parakeet model, streaming 160 ms chunks on the Neural Engine. A second, more accurate pass locks in each sentence as you pause, so the final text is ready about 50 to 75 ms after you let go.
How do I get it?
Murmur is in early access. Email me to join.