Rekk

A tiny macOS audio recorder that captures your microphone and the system audio at the same time, cancels the speaker → mic echo with WebRTC AEC, and transcribes the result locally with whisper.cpp. No cloud, no Python, no subscription.

What it does

• Mic + system audio in one WAV. ScreenCaptureKit taps the system output natively; no virtual audio driver, no BlackHole / Loopback dance.
• Acoustic Echo Cancellation (WebRTC AudioProcessing). Record meetings on speakers without the doubled-voice effect.
• Local transcription via whisper.cpp with Metal acceleration. Models stay on disk under ~/Library/Application Support/io.dbr.rekk/models/. Nothing leaves the machine.
• Streaming segments: transcript lines appear as Whisper decodes them, not at the end.
• Compact LCD-handheld UI that pins to a corner of your screen while you work.
• Pause / resume, per-source mute (mic / system), language selector (ES / EN / Auto), re-transcribe in another language at any time, files browser with reveal in Finder.

Requirements

• macOS 11+ (Apple Silicon recommended; Metal backend kicks in there).
• About 600 MB free for the default small Whisper model (other sizes available from inside the app).
• Microphone permission and Screen & System Audio Recording permission (the app walks you through them under setup).

Install

Download Rekk_<version>_aarch64.dmg from the Releases page, open it, drag Rekk.app to /Applications.

The binary is not notarized — first launch will need a right-click → Open → Open to dismiss Gatekeeper, or xattr -dr com.apple.quarantine /Applications/Rekk.app.

Build from source

# one-time toolchain (Homebrew)
brew install cmake meson ninja

# clone + run
git clone https://github.com/danibram/rekk
cd rekk
npm install
npm run tauri build     # → src-tauri/target/release/bundle/macos/Rekk.app
                        # → src-tauri/target/release/bundle/dmg/Rekk_*.dmg

The first build compiles whisper.cpp (with Metal) and webrtc-audio-processing from source — expect 2–4 minutes once, cached afterward.

For development hot-reload:

npm run tauri dev

Architecture

┌────────────────────────────────────────────────────────────┐
│                          React UI                          │
│  topbar · LCD strip · controls · drawer (files/tx/setup)   │
└────────────────────────┬───────────────────────────────────┘
                         │  Tauri IPC (invoke + event)
                         ▼
┌────────────────────────────────────────────────────────────┐
│                       Rust backend                         │
│                                                            │
│  CPAL ─┐   ScreenCaptureKit ─┐                             │
│        │                     │                             │
│        ▼                     ▼                             │
│   mic stream ───► WebRTC AEC ◄── system audio              │
│                       │                                    │
│                       ▼                                    │
│                  mixed WAV file ───► whisper.cpp (Metal)   │
│                                              │             │
│                                              ▼             │
│                                       .txt transcript      │
└────────────────────────────────────────────────────────────┘

Key crates: cpal, screencapturekit-rs, webrtc-audio-processing, whisper-rs, hound for WAV I/O, tauri v2 for the shell.

macOS permissions

On first record Rekk prompts for:

• Microphone — captures your voice.
• Screen & System Audio Recording — required by ScreenCaptureKit to tap the audio. Rekk never reads pixels; if you only want to grant the audio sub-permission, see below.

Granting only system-audio access (macOS 14.4+)

If you don't want Rekk to be able to see your screen, macOS 14.4 added a sub-permission specifically for this case. After the first prompt:

1. Open System Settings → Privacy & Security → Screen & System Audio Recording.
2. Scroll down to the System Audio Recording Only section.
3. Enable Rekk there and leave Rekk disabled in the upper "Screen & System Audio Recording" list.

A future migration to the Core Audio Process Tap API (NSAudioCaptureUsageDescription) would make this the default and skip the screen recording prompt entirely.

Phantom permissions after a rebuild

Every rebuild produces a binary with a new ad-hoc code signature, which macOS treats as a different app. The toggle in System Settings stays on but TCC denies access. Rekk's setup → reset & restart button runs tccutil reset for the bundle id and relaunches the process so macOS re-prompts cleanly.

Credits

• Georgi Gerganov for whisper.cpp and the GGML model format hosted on huggingface.co/ggerganov/whisper.cpp.
• The WebRTC team for the AudioProcessing module, repackaged by PulseAudio and then by tonarino for Rust.
• doom-fish/screencapturekit-rs for the safe Rust bindings to ScreenCaptureKit.
• tazz4843/whisper-rs for the whisper.cpp bindings.
• The Tauri project for the shell.

License

MIT · © 2026 danibram