Fast local dictation and call transcription for Mac.
Dictate types what you say into any app, and transcribes your online calls, marking what you said and what the others said. No bot joins the call.
What it does
Dictation in any app
Press Option ` and talk, and Dictate transcribes as you go. Press it again and the text lands where you were writing.
Calls are transcribed
Dictate notices when you're on a call and offers to transcribe it. It records your voice apart from the call with BlendKit, the audio engine from Blend Mic.
Unsure words get a second try
Whisper scores how sure it is of each part. Dictate transcribes the unsure parts again and keeps the better version. Fix the text before pasting lets a model on your Mac (Apple Intelligence or Ollama) correct misheard words.
Recordings delete themselves
By default Dictate deletes the audio after 7 days and the transcript after 30. You choose how long each is kept.
Your agents can read it
Claude Code or any MCP client can search your dictations and calls.
History
Every dictation, call and imported file in one place. Search it, correct a word or While the audio is kept. You choose how long in Data retention..
And more
- Add the names Whisper gets wrong as key terms, spelled your way.
- A level meter shows that the mic hears you.
- The text is ready almost as soon as you stop, because Dictate transcribes while you talk.
- Only the shortcut stops a recording, so a stray key never loses what you said.
- Your clipboard comes back after Dictate pastes.
- Your five newest recordings are a click away in the menu.
How it compares
Dictate
Free
Free forever for the first 50 early access users. After that, a lifetime license is $49. All processing happens on your Mac.
Local apps
Handy, OpenSuperWhisper, VoiceInk, Superwhisper and MacWhisper
Free
Some apps charge up to $75 once. Processing happens on your Mac.
Why I made it
I wanted local dictation and call transcription on my Mac that is fast, simple and robust. Local transcription on Apple silicon is so good now that Mostly. Local models still slip when you switch languages mid-sentence. They also do worse in languages Whisper knows less well. A noisy room trips up every model, cloud ones included. process your data in the cloud. Handy and OpenSuperWhisper are my all-time favorites for local dictation, and I've learned a lot from them.
I spent a lot of time using those tools and ran into a few problems. Sometimes I'd talk and the mic was off, so Dictate shows a level meter and your words as you talk. Sometimes I'd hit Escape by accident and lose what I'd said, so only Dictate's own shortcut stops a recording.
The biggest thing, though, was calls. I wanted the same setup to work for my online calls, reliably. I used to record them on my phone, which meant keeping the call on speaker. So I built Blend Mic, a virtual mic that also hears your Mac's sound. That let me pass a call into tools that already transcribe well, like dembrane, Handy and OpenSuperWhisper.
People kept asking me about the setup, mostly people who didn't want to download speech models and set up a virtual mic themselves. So I made the whole thing easy to set up and put it out there. Dictate has Blend Mic's audio engine built in, and no bot joins the call. Your agents can read all of it.
Questions
Should I ask before I transcribe a call?
Yes. Ask before you press Record, and don't record if they say no. People do mind, and with no bot in the call they only know if you tell them. You could say: "Can I transcribe this for my notes? It stays on my Mac and the audio is deleted within a week." How to record calls with consent.
For me it's personal. I'm not a native English speaker, and recording lets me stay present in the call instead of scrambling for notes. I take my notes afterwards.
By default the audio is deleted after 7 days and the transcript after 30. In Settings, Setup, you can delete the audio as soon as it's transcribed, or after 1 day.
What does it cost?
Free forever for the first 50 people. After that, it's $49. You pay once, with no subscription, unlike apps that run in the cloud.
Does any of my data leave my Mac?
No. Dictate turns your voice into text right on your Mac, so it works even without internet. It only goes online for two things:
- the first time you set it up, to download its speech recognition models (about 1.6 GB)
- now and then, to check for a new version. Nothing you said or wrote is sent.
Your recordings and their text stay on your Mac. Recordings are deleted after 7 days and text after 30, and you can change that in Settings, Setup. If you decide to connect an AI assistant such as Claude, it can read your text. What it does with it depends on that assistant, so pick one you trust. Where everything is kept
Does it work with Slack, Teams, Google Meet and WhatsApp?
Yes, and with Zoom, FaceTime, Discord, Signal and Telegram too. Calls in a browser tab work as well. How to transcribe calls on a Mac.
Which languages does it understand?
English works best. Spanish, French, German, Italian, Portuguese, Dutch, Japanese, Chinese and Korean work well too, and Dictate picks up the language on its own. Other languages work with more mistakes. For those, choose Large v3 in Settings, Model.
Will it run on my Mac?
Yes, if your Mac has an M1 chip or newer and macOS 15 or later.
What does it ask permission for?
Your microphone, to hear you. Accessibility, to paste the text into the app you're in. Call audio, to hear the other people on a call. You can skip call audio and still dictate.
Why do I need to activate it?
I want to know who uses Dictate. You enter your email once and get a code. I might write to ask how you use it and what's missing. Tell me to stop and I will. The app never signs in and never reports what you do.
Can an AI assistant read my transcripts?
Only if you decide to connect one, such as Claude. It can then read your dictations and calls. It can't record anything unless you allow it in Settings, Agents. Pick one you trust, because what it does with your text is up to it.
How is it different from other dictation apps?
Handy and OpenSuperWhisper are my favorites, and I learned a lot from both. Dictate adds three things. It listens again when it's unsure of a word. It transcribes your calls, marking what you said and what the others said. And an AI assistant can read it all, if you connect one.
Is it open source?
No. It is source available under FSL-1.1-MIT. You can read, change and run the code. You cannot offer it in a commercial product that competes with it. Each version becomes MIT two years after release, and versions up to 0.11.0 already are. The code and license are on GitHub.
Can I use it on my phone?
Not yet. I record the call as a voice memo on my Apple Watch. Then I import the file into Dictate on my Mac.
How do I send feedback?
In Dictate, choose Send Feedback… in the menu bar menu or in Settings, About. A short form opens in your browser with your activation ID and versions filled in, so I can find what went wrong. On older versions, or without Dictate, use the feedback page. For a bug, say what you did and what happened instead. I read every message, and I reply when you leave an email.
