SwarmPane/Voice/Pro · macOS
Too much to type? Say it.
Press ⌘⇧Space and talk: the words land as a draft in your agent’s composer or terminal, and nothing reaches the agent until you send it.
Mac (macOS 13 or later) and Windows, with Linux next · Pro $29/month
SwarmPane VoiceThe shortcut
Press once to talk, again to insert. Esc cancels.
- Cloud
- Deepgram Nova-3, with Whisper large v3 turbo as backupCloud voice, run on Cloudflare
- Minutes
- 1,500 minutes a monthAbout 25 hours of cloud voice, with Pro
- Offline
- Whisper on your MacWorks offline and uses none of those minutes
Read what happens
SwarmPane is open in Agent mode on a conversation with Release partner. Three keys, ⌘, ⇧ and Space, go down, and the capsule above the composer says Listening, with the one-time notice “Voice uses SwarmPane cloud. Audio is never stored.” The film has no sound, so the sentence appears beside the app as it is said, over its waveform: “Add a loading state to the checkout button and keep every test green.” The same keys finish it. After Transcribing, the words land in the composer as a draft, and the capsule says “Added to your message. Edit it, then send.” A few typed words are added (“Then show me the diff.”) and Return sends the message. A second dictation starts, “Also rename the pay button to…”, and Esc cancels it: nothing is inserted. In Code mode the toolbar’s Voice button dictates into the claude pane: “Run the checkout tests and show me only the failures” is typed on its input line and not run, and the bar says to review it, then press Return. Then the two engines: SwarmPane cloud, the default, runs Whisper large v3 turbo on Cloudflare Workers AI, where the audio is transcribed, then dropped; Whisper on your Mac keeps the audio on it. Offline, the capsule says “You’re offline; using Whisper Base on this Mac.” The film ends on SwarmPane Voice: Say the change. Send it when it’s right. With Pro, on macOS 13 or later.
01/05Try itSimulated · no microphone
Press the keys. See where it lands.
The app’s composer, terminal, Voice Anywhere and Settings, rebuilt with sample data. ⌘⇧Space works here while the demo is on screen, and Esc cancels.
Try it in your browser
Simulated dictation with sample text. No microphone, no uploads. Press or the microphone to dictate a sample.
02/05Where it types3 places
Hands on the keyboard? Keep them there.
⌘⇧Space starts a dictation and the same keys insert it. Esc cancels. It works from any app: the words go to the composer or terminal that is active in SwarmPane.
In a composer
Agent, Chat, Workflow, Video Studio and Point and Fix: the words arrive as a draft. Edit it, then send.
In a terminal
Code mode types the words on the pane’s input line and never presses Return. You read it, then run it.
In any Mac appLabs
Voice Anywhere puts the words in the app in front instead. It never types into password fields or presses Return, and ⌥ ⇧ ⌘ Space speaks commands to SwarmPane.
Voice Anywhere is in Labs and off by default while its last checks finish. It needs macOS Accessibility permission.
03/05Engines2 engines
Offline or out of minutes? It falls back to your Mac.
SwarmPane cloud is the default. When it can’t be used, dictation runs on your Mac with Whisper, and the app tells you why.
Deepgram Nova-3 and Whisper large v3 turbo are multilingual and detect the language as you speak. Whisper Base and Whisper Turbo are multilingual too, but SwarmPane runs them in English for now. Our own measurements so far are in English.
Cloud voice is billed by the second. Minutes reset every month and don’t roll over.
04/05PrivacyAudio never stored
Where does the audio go? Nowhere it’s kept.
Read the privacy policySwarmPane cloudDefault
- Your MacRecords while you talk, and sends it when you finish
- SwarmPane’s serviceDeepgram Nova-3, or Whisper large v3 turbo when Nova-3 is unsure, transcribes it on Cloudflare Workers AI
- Your MacThe words arrive as a draft
Kept: the seconds you used, and each request’s length, timing and outcomeNot kept: the audio, the text
On this MacWorks offline
- Your MacRecords while you talk; Whisper transcribes it when you finish
- Your MacThe words arrive as a draft
Never leaves your Mac
Transcribed, then dropped.
Cloud voice sends each recording to SwarmPane’s service, where Deepgram Nova-3 transcribes it on Cloudflare Workers AI (Whisper large v3 turbo takes any clip Nova-3 is unsure of). Neither the audio nor the text is stored or logged.
Not used for training.
Cloudflare does not use the audio or the text to train models.
Counted, never read.
To count minutes the service keeps the seconds you used. Each request’s log has its length, timing and outcome, never its words.
Or keep it on your Mac.
With Whisper on your Mac the audio never leaves it. The first cloud dictation says “Voice uses SwarmPane cloud. Audio is never stored.” and offers on-device voice in one click.
05/05Questions8 answers
Questions, answered.
More in the voice guide and the privacy policy.
Pro. It includes 1,500 minutes of cloud voice a month, about 25 hours (300 in the first 7 days of a first Pro plan), billed by the second; minutes reset every month without rolling over. Dictation with Whisper on your Mac uses none of them. Starter has no voice, on the Mac or in the cloud.
If Whisper is set up on your Mac, dictation runs there and the app says why. If it isn’t, the app says what is missing and offers the fix: download Whisper Base (148 MB) or install whisper.cpp.
No. With cloud voice the recording is transcribed and dropped: neither the audio nor the text is stored or logged, and Cloudflare does not use them to train models. With Whisper on your Mac, the audio never leaves it.
Cloud voice uses Deepgram Nova-3, with Whisper large v3 turbo as backup; both are multilingual and detect the language you speak. On your Mac, SwarmPane runs Whisper in English for now. Our own measurements so far are in English.
No. In Code mode the words are typed on the pane’s input line and never run. You read them and press Return yourself.
The shortcut works from any app and sends the words to SwarmPane. Voice Anywhere, in Labs, types them into the app in front instead. It needs macOS Accessibility permission, never types into password fields and never presses Return. It is off by default while its last checks finish.
Cloud voice takes clips up to 2 minutes. A longer recording is transcribed on your Mac instead, if Whisper is set up there.
SwarmPane Voice runs in the Mac app (macOS 13 or later) and the Windows app, where the shortcut is Ctrl+Shift+Space. SwarmPane itself runs on Mac, Windows and Linux; the Linux app gets voice next. Offline Whisper is in the Mac app only.
→StartWith Pro · macOS 13 or later
Say the change. Send it when it’s right.
SwarmPane Voice comes with Pro, along with Video Studio and 10 million Jev tokens.
Pro $29/month · Compare Starter and Pro
Cloud voice, by plan1 cell = 1 hour
- Pro
- 1,500 minutes a month, about 25 hours
- Trying Pro
- 300 minutes during a 5-day Pro trial, or a first Pro plan’s first 7 days
- Starter
- No voice, on the Mac or in the cloud
Whisper on your Mac uses none of these minutes. Cloud voice is billed by the second, and minutes don’t roll over.