Biscotti
Private Meeting Transcripts for MacOS.
Biscotti is a private and free meeting recorder for macOS. It records your meetings and turns them into accurate, speaker‑labeled transcripts and summaries. No cloud. No bot joining your call. No subscription.
BiscottiDemoHQ.mp4
Features
- 🔒 Private by design — recording, transcription, and AI all run on your Mac. Your data stays local.
- 🆓 Totally free — no account, no subscription, no upsell.
- 🧠 Powerful AI summaries - automatic summaries, action items, meeting titles, and real speaker names.
- 🤖 No bots, any app — records Zoom, Teams, Meet, FaceTime, or Slack huddles — even an in‑person conversation — without joining your call.
- 🗣️ Speaker identification — accurate transcripts, automatically split by speaker and tagged with speaker name.
- 📅 Calendar‑aware — sees your upcoming meetings, offers to start recording.
- 🧩 MCP - chat with your meeting notes in apps like Claude Desktop or LM Studio.
- ⏹️ Auto‑stop — detects when your call ends and stops recording.
- 📖 Custom vocabulary — transcribe uncommon words you use, like names or jargon.
- ⚡ Fast, small, native — launches instantly, native design, built by an ex‑Apple engineer.
- 🎤 Voice isolation — captures your mic and everyone else as separate, clean channels. No echo.
- 📝 Markdown notes — jot notes in markdown, linked to the moment they happened.
Biscotti vs. other notetaking apps
| Biscotti | Other notetaking apps | |
|---|---|---|
| Audio never leaves your Mac | ✅ | ❌ |
| Meeting notes stay private and local | ✅ | ❌ |
| No bot joins your meeting | ✅ | ❌ |
| Works with any app — even in person | ✅ | ❌ |
| Keeps full audio, not just summaries | ✅ | ❌ |
| Native Mac app, not Electron/web | ✅ | ❌ |
| Free, no subscription, no account | ✅ | ❌ |
| Never have to pay to read old meetings | ✅ | ❌ |
| Where your data lives | Your Mac | Their Cloud |
How it works
- Record — Biscotti captures your mic and your Mac's system audio from any meeting app.
- Transcribe — on‑device speech recognition produces a speaker‑labeled transcript, powered by Apple Silicon.
- Understand — local AI writes a summary, pulls out action items, and figures out who's who.
Every step happens on your Mac. Your audio and transcripts never leave your Mac.
Requirements
- A Mac with Apple Silicon (M1 or later)
- macOS 15 (Sequoia) or later
- 16GB of RAM recommended
Install
Direct Download
- Download
Biscotti.dmg. - Open the dmg, then copy Biscotti.app to your Applications folder.
- Open Biscotti and follow the quick setup to grant permissions for microphone, system‑audio, and calendar access.
Homebrew
If you use Homebrew, installation is easy:
- Run
brew install biscotti - Now, open Biscotti and follow the above instructions as you normally would.
FAQ
Is it really private? Yes. Recording, transcription, and AI all run locally on your Mac. There's no account, and your audio and transcripts are never uploaded. It works completely offline (after downloading models).
Does it join my meeting as a bot? No. Biscotti records your Mac's audio directly, so nothing appears in the call and other participants aren't notified.
What apps does it work with? Anything that plays or records audio — Zoom, Microsoft Teams, Google Meet, FaceTime, Slack, Webex — plus in‑person conversations through your microphone.
Does it cost anything? No. Biscotti is free, with no subscription and no account.
Where is my data stored? Locally, on your Mac.
Which AI does it use? All AI runs locally on your Mac.
- Transcription powered by Whisper V3 Turbo, run by WhisperKit
- Speaker identification by Pyannote, run by SpeakerKit
- Language models by Google Gemma 4, run on llama.cpp
How does it know when to suggest starting/stopping recording? Biscotti monitors active audio apps, without listening to their audio streams. Nothing is recorded unless you click record.
Can I connect my calendars? Yes. Biscotti can sync any calendars you connect to the Apple Calendar app on your Mac. Event data enables meeting start notifications, speaker identity matching, and enhances meeting summaries.
How does speaker idenitification work? Four steps: 1) an AI model transcribes what you say, 2) a second AI model separates speakers by voice + generates voiceprint vectors, 3) we search for voiceprints closely matching these voiceprints in your past meetings which have speaker labels, and 4) a third model figures out who's who balancing evidence (voiceprints, user said "Hi, I'm Steve", etc). Enhanced by calendar metadata when available.
Which languages does it support? The transcription model supports 99 languages and summary model supports 35+ languages.
Built with
On‑device speech recognition and speaker identification powered by WhisperKit and SpeakerKit from Argmax.