MeetRec

Record meetings, livestreams, and calls on macOS — locally, into a single stereo M4A file. Left channel is system audio, right channel is your microphone. No cloud. No bot in your meeting. No virtual audio driver. Works with AirPods.

Demo gif coming soon — menu bar icon, hotkey muting mic, file appearing in Finder.

Install

git clone https://github.com/BozhengLong/meetrec.git
cd meetrec
./make-app.sh

Drag MeetRec.app to /Applications and double-click. On first launch macOS will block it (ad-hoc signed), so either:

• System Settings → Privacy & Security → Open Anyway, or
• xattr -dr com.apple.quarantine /Applications/MeetRec.app

Or grab the prebuilt zip from Releases and do the same.

What it does

• 🎙️ Captures system audio (via ScreenCaptureKit) + microphone (via AVAudioEngine) simultaneously
• 🎚️ Stereo M4A output: L = system, R = mic — separated for clean transcription downstream
• 🔇 Independent per-channel mute, real-time, with global hotkeys
• 📊 Live level meters in the popover so you can see audio is actually flowing
• ⏱️ Recording time visible right in the menu bar (no need to click)
• 🎧 Works with AirPods and other Bluetooth output, unlike Process-Tap–based tools
• 💾 Saves to ~/Recordings/YYYY-MM-DD_HHMM.m4a, opens Finder on stop

Hotkeys

Combo	Action
⌥⌘R	Start / stop recording
⌥⌘M	Toggle microphone
⌥⌘S	Toggle system audio

How it compares

	MeetRec	Granola	Audio Hijack	BlackHole + OBS
Price	Free	Subscription	$64 one-time	Free
Local-only	✅	❌ (cloud transcription)	✅	✅
Joins meeting as bot	✅ no	✅ no	✅ no	✅ no
Needs virtual audio driver	✅ no	✅ no	✅ no	❌ yes
Works with AirPods	✅	✅	✅	✅
Built-in transcription	❌ (BYO Whisper)	✅	❌	❌
Open source	✅	❌	❌	✅ (BlackHole only)
Setup complexity	one script	install	install	multi-step routing

If you want auto-summarized meeting notes with live transcript, use Granola. If you want a full audio production suite, use Audio Hijack. If you just want a clean local recording that you'll feed into Whisper/ChatGPT yourself, that's what MeetRec is.

After recording

MeetRec deliberately stops at "make a clean audio file." Transcribe with whatever you prefer:

# Split L (system) and R (mic) into two mono files for cleaner transcription
ffmpeg -i 2026-05-24_1430.m4a \
  -map_channel 0.0.0 system.m4a \
  -map_channel 0.0.1 mic.m4a

# Then transcribe each side
whisper-cli -m models/ggml-large-v3.bin -f system.wav -l zh

Or just drag the m4a into MacWhisper, Buzz, or ChatGPT.

Permissions you'll be asked for

Permission	Why
Microphone	Records your voice into the right channel
Screen Recording	ScreenCaptureKit needs it to capture system audio. No video is ever saved — the dummy 2×2 video stream is discarded

If you only see the mic side recorded and ~/Recordings/meetrec.log contains SCShareableContent FAILED ... declined TCCs, the Screen Recording permission is denied. Open System Settings → Privacy & Security → Screen Recording, enable MeetRec, then fully quit and relaunch. If MeetRec doesn't appear in the list at all (macOS suppressing the prompt after a previous denial), reset and try again:

tccutil reset ScreenCapture com.local.meetrec

This is most likely to happen after rebuilding from source — re-signing changes the app identity and TCC may silently remember a stale "denied" state.

Known limitations

• No transcription — by design. Feed the m4a to whatever ASR you prefer.
• No pause within a single file — stop & restart creates a new file.
• Switching the system output device mid-recording is untested. Start playback first, then start recording.
• Ad-hoc signed, not Apple-notarized. First launch needs a Gatekeeper override (see Install above).

Requirements

• macOS 14.2 or later
• Apple Silicon or Intel Mac
• Xcode 15+ / Swift 5.9+ toolchain (for building from source)

Build from source

git clone https://github.com/BozhengLong/meetrec.git
cd meetrec
./make-app.sh         # → ./MeetRec.app
# or, just the binary:
swift build -c release

Architecture

┌──────────────────────────────────┐
│  SCStream (capturesAudio=true)   │  ← system audio
│  AVAudioEngine.inputNode         │  ← microphone
│      ↓ (independent callbacks)   │
│  StereoWriter                    │  ← heartbeat-flushed every 100ms,
│      ↓                           │     pads silence for the muted side
│  AVAssetWriter → AAC m4a stereo  │
└──────────────────────────────────┘

• SystemAudioCapture — wraps SCStream with capturesAudio=true and a minimal 2×2 video stream that's discarded inside SCKit.
• MicrophoneCapture — AVAudioEngine.inputNode with a tap, requests permission explicitly via AVCaptureDevice.requestAccess.
• StereoWriter — own dispatch queue, two mono Float32 ring buffers, every 100ms emits a stereo interleaved chunk through AVAssetWriter. Mute = append zeros instead of samples; the timeline stays aligned, no click/pop.
• HotKeyManager — Carbon RegisterEventHotKey, no dependencies.
• AudioLevel — peak-detection over PCM buffers for the level meters.

File layout

meetrec/
├── Package.swift
├── make-app.sh                   # builds + packages MeetRec.app
├── LICENSE
├── README.md
└── Sources/MeetRec/
    ├── MeetRecApp.swift          # app entry, MenuBarExtra, hotkey wiring
    ├── MenuBarView.swift         # SwiftUI popover with level meters
    ├── RecordingEngine.swift     # orchestrator (ObservableObject)
    ├── SystemAudioCapture.swift  # ScreenCaptureKit-based system audio
    ├── MicrophoneCapture.swift   # AVAudioEngine
    ├── StereoWriter.swift        # AVAssetWriter + L/R interleave
    ├── HotKeyManager.swift       # Carbon global hotkeys
    ├── AudioLevel.swift          # peak detection for meters
    └── Log.swift                 # file-based logger

Why not Core Audio Process Taps?

The Process Tap API (AudioHardwareCreateProcessTap, macOS 14.2+) is the "obvious" way to capture system audio. We tried it first. It works on built-in speakers and most output devices — but silently fails when the default output is Bluetooth in HFP/SCO mode (AirPods etc. while your app is also recording from the mic). In that case macOS switches the Bluetooth profile to the lower-bandwidth call mode, and the tap's IO proc never fires — AudioDeviceStart returns noErr but no callbacks come in. ScreenCaptureKit's audio path sits above CoreAudio's device routing and doesn't have this issue.

License

MIT — see LICENSE.