Hey, after every call you’ve got the minutes ready — with names, who said what. Are you typing all that up by hand?
No, that’s VoiceGrep. It sits in the menu bar and notices on its own when a call starts — Zoom, Teams, Google Meet, Telegram, whatever. I don’t press a thing.
It records two tracks — system audio and the microphone. The call ends, the recording stops, and right there on the Mac it all turns into text.
Hold on. “Turns into text” — where, exactly? These are client calls. We can’t hand them over to someone else’s cloud.
They’re not handed over anywhere. Recording, transcription and voice recognition all run on the device. That’s not a checkbox you could untick by accident — it’s how the app is built.
Does it need the internet?
Once — to download the speech model, about 0.6–1 GB. After that it all works offline. The network is only for updates, and for an external LLM if you turn one on with your own key.
Does it handle English? Most of our calls are in English.
Yes — English and Russian. Russian is the default, so switch the meeting language to English in Settings, or let it auto-detect. Heads-up: the interface itself is in Russian for now, and so are summaries and tasks — though the assistant can answer in English.
Wait, how does it tell my voice from yours?
You’re the microphone, so it labels you “Me”. Everyone else it tells apart by voice: name someone once, their voiceprint is saved, and from then on they’re labeled automatically in every meeting. It already knows Olivia from earlier calls. Me it doesn’t know yet — name me, there’s a button above, and watch what happens to the lines.See — my name is already on every line, above and below.
What if calls run back to back? Some days I have five in a row.
When it records automatically, it usually splits them into separate meetings on its own. And it tries to capture only the call app’s audio, so notifications, music and random pings from other apps stay out.
What permissions does it ask for? Screen Recording makes me nervous.
Microphone and Screen Recording. Screen Recording is just how macOS lets it capture the call’s system audio — no screen video is ever saved.
Okay, so it’s recorded. Then what? Nobody rereads transcripts.
You don’t reread them. Each meeting has search, and for the whole archive there’s the assistant: ask “what did we agree with Sam?” or “what are my tasks?” — it searches the transcripts itself, even by sound, and answers with links back to the conversations. The assistant needs an external LLM on your own key — that’s opt-in.
Does it do summaries and tasks too?
If you connect an LLM with your own key — yes: a short summary and tasks with owners. It only suggests tasks; you confirm the ones you want. There’s also a card for every person — role, company, notes — and a graph of who meets with whom.
Is the graph just a picture?
No, it draws conclusions. You see where your time goes — by circle of people or by project, who connects different teams and which of them only talk through you, whom you meet less often and who has dropped off entirely. All computed on your Mac, from your own meetings.
Can I mark something during the call? An important moment, say.
There’s a floating panel you can turn on in Settings: a timer, marks — important, question, task, decision — and screenshots. Everything is pinned to its moment in the conversation. Try it — the mark lands right on the timeline.
And if the Mac crashes in the middle of a call?
The meeting exists from the first second of recording. After a crash, the recording is recovered and transcribed. And it backs up its database regularly on its own.
What about recordings I already have?
Sure — import an existing recording, audio or video, and it’s transcribed the same way, with names, a summary and tasks. You can export any meeting to Markdown, and move voiceprints to another Mac of yours.
Sold. How do I install it?
Download it, drag it into Applications and launch it. It’s signed and notarized by Apple — no Terminal commands, and after that it updates itself.
Which Mac do I need?
Apple Silicon — M1 or later, on macOS 14.4 or newer. It’s about 23 MB. And it’s free — recording just needs an activation key: copy the Machine ID from Settings, message the developer on Telegram, and you get a key.
Done — installing it.
Call ended · recording stopped on its own · 05:12
Meeting recap
This is what the meeting card looks like after a call when an external LLM is connected (optional, on your own key): it writes the summary and suggests tasks. Transcription and names work without it.
Summary
- VoiceGrep notices a call in Zoom, Teams, Google Meet or Telegram, records system audio and the microphone, and transcribes both right on the Mac after the call — no cloud.
- Transcribes English and Russian. Russian is the default — switch the meeting language to English. The interface is in Russian for now.
- Name a person once — from then on they’re recognized by voice in every meeting.
- Search inside each meeting and an assistant across the whole archive (via an external LLM), marks and screenshots during the call, importing your own recordings, Markdown export; a people graph — circles of people, where your time goes by circle and by project, who connects teams and whom you meet less often.
- Apple Silicon, macOS 14.4+, about 23 MB, free; recording needs an activation key, issued by Machine ID via Telegram.
Tasks4
-
Download VoiceGrep
-
Switch the meeting language to English
-
Name the new speaker once
-
Get an activation key
VoiceGrep on your phone
An Android beta: a remote for the archive on your Mac. The Mac stays in charge — it stores and processes your meetings, while the phone shows them and sends edits and recordings back to the Mac.
What you can do from the phone
- Search meeting text, read transcripts, recaps and tasks, play back lines.
- Name an unnamed voice or move lines to another person — voice recognition remembers it.
- Record an in-room meeting: the recording goes to your Mac and is transcribed there like any call.
- Ask the assistant about your meetings and get a daily recap (needs an LLM set up on the Mac).
How to connect
- On the Mac: Settings → Данные → Телефон (Data → Phone), turn on Доступ с телефона (Phone access). It’s off by default.
- Click Подключить телефон… (Connect phone…) and scan the QR code with the app. The first time, the phone and the Mac must be on the same Wi-Fi.
- Away from home the phone connects through a relay, which only passes encrypted data. The Mac has to be on.
- Android 10+. An iPhone version is coming later. Heads-up: the app is in Russian.
For automatic updates use Obtainium: add github.com/ilyabazhenov/voicegrep-android there.
Questions and answers
The same things the call covered — short and to the point.
What is VoiceGrep?
A Mac app that lives in the menu bar: it records your online calls on its own, transcribes them on-device once the call ends, and labels each line with the speaker’s name by recognizing their voice. Recordings and transcripts never leave your Mac, and no bot joins your call.
Which calling apps does VoiceGrep work with?
Any app that uses your Mac’s microphone for the call: Zoom, Microsoft Teams, Google Meet, Telegram, FaceTime, Slack, Yandex Telemost and others — including calls in the browser. VoiceGrep notices when a call starts and stops recording when it ends. When it records automatically, it tries to capture only the calling app’s audio, keeping notifications and music from other apps out of the recording.
How do I transcribe a Zoom or Teams call on my Mac?
There’s nothing special to do. Install VoiceGrep, allow Microphone and Screen Recording access, enter your activation key, and just take your calls. When a call ends, the transcript appears in the meeting card — with timestamps, speaker labels and search. No meeting links to paste and no bot to invite.
Where do my recordings and transcripts go?
They stay on your Mac: audio, text and voiceprints are stored locally, and speech and voice recognition run on-device. It isn’t a setting you could flip by accident — it’s how the app is built.
The network is only used to download the speech model once (0.6–1 GB) and to check for updates. If you connect a cloud language model for summaries and the assistant, it receives meeting text (transcripts, notes, attachments) — never audio or voices; with a local model (Ollama, LM Studio) not even that leaves your Mac.
Do I need to tell people I’m recording?
Yes. Say so at the start of the call and get consent where the law or your company requires it. VoiceGrep doesn’t join the meeting, so the call service won’t tell participants about the recording for you.
Is it free? What’s the activation key?
VoiceGrep is free. To record or import new meetings you need an activation key: it’s tied to your computer and issued for a set period. Open Settings → Лицензия (License), copy the ID машины (Machine ID) and message us on Telegram — we’ll send you a key.
The app reminds you 14 days before the key expires; a new Mac needs a new key. Without a key, everything already recorded stays available: listen, search, export.
Is there a phone app?
There’s an Android beta (10+): meetings, transcripts, sorting out who spoke, the assistant and recording in-room meetings. It pairs with your Mac — turn on Settings → Данные → Телефон → Доступ с телефона (Phone access) there and connect the phone with a QR code. More ↑
An iPhone version is coming later.
Which languages does VoiceGrep transcribe?
English and Russian. Russian is the default, so for English calls switch the meeting language to English in Settings, or turn on auto-detect. The interface itself is in Russian for now, and so are summaries and tasks.
How does VoiceGrep know who’s speaking?
Your voice comes from the microphone, so your lines are labeled “Me”. Everyone else is told apart by voice: name a person once, their voiceprint is saved, and VoiceGrep labels their lines on its own in future meetings.
Can it transcribe a recording I already have?
Yes. Import audio or video (m4a, mp3, wav, mp4, mov and other formats macOS can open) and VoiceGrep transcribes it just like its own calls: split by speaker, plus a summary and tasks if a language model is connected. Any meeting can be exported to Markdown.
Does VoiceGrep write summaries and action items?
Yes, once you connect a language model: with your own API key, through your Claude or ChatGPT subscription (via their command-line tools claude and codex), or a local one (Ollama, LM Studio). After each call you get a short summary and tasks with owners — VoiceGrep suggests tasks, you confirm the ones you want. An assistant answers questions across your whole archive, with links to the conversations.
Transcription and speaker names work without any language model.
What does the people graph show?
It’s a separate window under People. The graph finds circles of people who meet together on its own and labels them with a project or company. Other tabs show how many hours a week go to each circle or project, who connects different teams and which of them only talk through you, and whom you meet less often or not at all anymore.
Everything is computed on your Mac from your own archive — no language model needed.
How is VoiceGrep different from cloud transcription services?
Cloud services process the recording on their servers, and many add a bot to your call to do it. VoiceGrep adds nothing to the call and uploads nothing: transcription happens on your Mac. That makes it fit for client calls you can’t hand to third parties (as long as you don’t connect a cloud language model) — and it works offline.
What are the system requirements?
A Mac with Apple Silicon (M1 or later) running macOS 14.4 or newer. The app is about 23 MB; the speech model is 0.6–1 GB and downloads once. It needs Microphone and Screen Recording permissions — the latter only so macOS hands over the call’s audio; no screen video is saved.