ScribeBuddy

macOS meeting transcription tool. Captures Zoom, Teams, Google Meet, and any browser tab in real-time using a local Whisper model — no data leaves your machine.

Requirements

• macOS 15 (Sequoia) or later
• Xcode 16+ Command Line Tools — xcode-select --install
• Rust — rustup.rs
• Tauri CLI — cargo install tauri-cli --version "^2.0"

Building

Step 1 — Clone and build

git clone https://github.com/mkarvan/ScribeBuddy.git
cd ScribeBuddy
./build.sh

On the first run, build.sh automatically creates a local code-signing certificate in your Keychain so macOS retains the Screen Recording permission across rebuilds. macOS will ask for your login password once to set the trust — that's normal and only happens this one time.

After that, every subsequent ./build.sh builds silently with no prompts.

To build and copy straight to /Applications:

./build.sh --install

Step 2 — First launch

The app is signed with a local certificate, not notarized by Apple, so Gatekeeper blocks it the first time:

1. Double-click ScribeBuddy.app — dismiss the "cannot be opened" dialog
2. Open System Settings → Privacy & Security
3. Scroll down and click Open Anyway

This only happens once.

Step 3 — Grant permissions

On first launch macOS will ask for:

• Microphone — to capture your voice
• Screen & System Audio Recording — to capture meeting audio (no screen content is recorded)

Click Allow for both.

Step 4 — Download a Whisper model

1. In the settings bar, choose a Model size (small is a good default — 461 MB, ~4 s latency)
2. Click the ↙ Download button and wait for it to complete

Models are stored in ~/Library/Application Support/ai.scribebuddy.app/ and reused across launches.

Development

# Hot-reload dev window (no signing needed)
cargo tauri dev

# Run the test suite
cargo test -p scribebuddy-core

# Type-check without a full build
cargo check --workspace

The test suite runs automatically before every ./build.sh via beforeBuildCommand in tauri.conf.json.

Troubleshooting

Screen Recording permission resets after every rebuild

This means the app was built without the signing certificate (e.g. with cargo tauri build directly instead of ./build.sh). Always use ./build.sh — it signs the app with a stable local identity so macOS retains the permission.

If the permission is stuck in a denied state, reset it and re-grant:

tccutil reset ScreenCapture ai.scribebuddy.app

Then relaunch and click Allow when prompted.

App isn't listed in Screen & System Audio Recording settings

Start the app first (so macOS registers it), then open System Settings → Privacy & Security → Screen & System Audio Recording.

Microphone not captured

Open System Settings → Privacy & Security → Microphone and confirm ScribeBuddy is enabled.

"Apple cannot verify" on every launch

You're running the app without signing it first. Always use ./build.sh — it signs the app automatically.

Architecture

[Microphone]   ──► cpal (ring buffer) ──┐
                                        ├──► AudioProcessor ──► Transcriber ──► Transcript
[Meeting App]  ──► ScreenCaptureKit ────┘        │
                   (48 kHz stereo → mono)    rubato resampler
                                             (source Hz → 16 kHz)
                                                  │
                                             WhisperEngine
                                             (Metal / CPU)

• AudioProcessor::run_with<T: Transcriber> is generic over the transcription engine — WhisperEngine in production, MockTranscriber in tests
• Session lifecycle (Idle → Recording → Paused → Stopped) is owned by SessionManager; Tauri commands are thin wrappers
• All file writes use write-to-temp-then-rename for crash safety

Stack

Crate	Purpose
whisper-rs 0.16	Whisper.cpp bindings with Metal acceleration
screencapturekit 0.3	macOS system audio capture
cpal 0.15	Microphone capture
rubato 0.14	Audio resampling to 16 kHz
ringbuf 0.4	Lock-free SPSC ring buffers
Tauri 2	App framework, system tray, native dialogs

License

MIT