Talk, and clean text appears where you're writing. Say what needs doing, and it becomes tasks in the tools you already use. Record a meeting and get the note. Speech never leaves your device unless you say so.
macOS 14 or later, Apple silicon. Free. No account needed.
One app, one hold key on the Mac, one mic button on the iPhone. The mode decides what happens to the words.
Hold, speak, release. Fillers and false starts are gone, sentences are fixed, and the tone matches where you're writing — casual in Slack, formal in Mail, verbatim in the terminal.
"um so can we uh move the call to thursday, actually no friday"Select some text, hold the edit key, say how to change it. The edited version replaces the selection.
"make it friendlier and change Tuesday to Thursday"Explain something out loud for two minutes. Review a short, ordered version before it goes anywhere.
Five minutes of thinking → a brief you'd sendSay what needs doing. Dates, tags and people stay with the task they were said with. Review, then copies go to Reminders, Obsidian, Microsoft To Do, ClickUp or Notion.
"call John about the invoice Friday, and ping design for the mockups"Record a call or a room. Speakers are told apart, each turn transcribed, and you get a Markdown note with a summary. On the Mac it hears both sides of the call.
Note lands in ~/Documents/VoiceMerge MeetingsModes are keyed on the app — and on the Mac, on the site too. Formal in mail.google.com, technical on a GitHub pull request, your own prompt anywhere you like. Your names and terms come along in a personal dictionary.
Speech-to-text always runs on the device. For the cleanup pass you pick one of three paths, and every dictation shows which one ran.
Parakeet or Whisper for speech, a small local model for cleanup — on the Mac's Neural Engine, or Apple Intelligence on the iPhone. Works on a plane.
Already have an Anthropic, OpenAI or Groq account? Paste a key. It's stored in the Keychain and the app talks to them directly.
Sign in with Apple, buy a credit pack, and cleanup, editing, Distill and meeting summaries run on Claude. You pay for what you use, and the ledger shows every request.
A menu bar app. Hold a key anywhere — any app, any text field — and the text is typed at the cursor when you let go.
Tap, speak, stop. The clean text is copied — or inserted into any app through the VoiceMerge keyboard.
Coming to the App Store. The keyboard needs Full Access to receive text from the app; it has no network code.
Everything on the device. No account, no subscription, no trial that ends.
Claude through VoiceMerge. A typical dictation costs a fraction of a cent; a meeting summary a few cents.
Not unless you choose it. Speech-to-text is always local. Cleanup is local by default. If you add your own API key or buy credits, the text of a request goes to that provider — never the audio, never the screen context.
Good for ordinary dictation: fillers, punctuation, sentence fixes, matching tone. It's a small model, so long Distill and meeting summaries are where Claude is noticeably better. The app tells you which path ran on every dictation so you can judge for yourself.
Most people never need the cloud tier, and those who do use it unevenly. Credits mean a $10 pack lasts months for light use, and the ledger shows exactly where it went. You can turn on auto-refill if you'd rather not think about it.
They stay in a history file on your device. Search it, re-clean an entry, copy the raw words, delete one or all. Turn history off and nothing is kept.
Parakeet handles English and 24 other European languages. Whisper covers around 99. Cleanup follows the language you spoke.
The apps and the API are on GitHub. The credit service is the part we run so you don't have to manage keys.