Ohh Brother

Passive meeting transcription app for macOS. Runs entirely local - no cloud transcription, no account required.

Features

• Local transcription using faster-whisper (no data leaves your machine)
• Menu bar app - start/stop recording with a click
• Streaming transcripts - markdown updates in real-time as you speak
• LLM summarization (optional) - summarize meetings with Claude or GPT

Requirements

• macOS 12+ (Apple Silicon or Intel)
• Python 3.11
• Microphone access
• ~500MB disk space for ML models

Installation

From Homebrew

brew install --cask https://raw.githubusercontent.com/tomfuertes/ohh-brother/main/Casks/ohh-brother.rb

Manual Installation

1. Download the latest release from GitHub
2. Move Ohh Brother.app to /Applications

Post-Installation Setup

The Python ML environment requires one-time setup:

# Navigate to app resources
cd "/Applications/Ohh Brother.app/Contents/Resources/python"

# Create and activate virtual environment
python3.11 -m venv venv
source venv/bin/activate

# Install dependencies (downloads ~500MB of ML models)
pip install -r requirements.txt

Usage

1. Click the microphone icon in your menu bar
2. Click "Start Recording"
3. Speak - the app captures audio and transcribes in the background
4. Click "Stop Recording" when done
5. Find your transcript in History > [date/time]

Summarization

1. Open Settings and add your Claude or OpenAI API key
2. After recording, click "Summarize Current"
3. The summary appears in a new window

Transcript Format

Transcripts are saved as Markdown and stream in real-time:

# Meeting - 2026-02-01 10:30 AM

[00:05] Let's get started with the Q3 review.
[00:15] Sure, I have the numbers here...

Development

Prerequisites

• Bun (JavaScript runtime)
• Python 3.11
• Node.js (for Electron)

Setup

# Clone repo
git clone https://github.com/tomfuertes/ohh-brother
cd ohh-brother

# Install JS dependencies
bun install

# Setup Python environment
cd python
python3.11 -m venv venv
./venv/bin/pip install -r requirements.txt
cd ..

# Run in development
bun run dev

Building

# Build Electron app
bun run build

# Package for distribution
bun run package

Architecture

┌─────────────────────────────────────────────┐
│  Electron Menu Bar App (Bun runtime)        │
│  - Menu UI (start/stop, history, settings)  │
│  - LLM API calls for summarization          │
│  - Spawns/manages Python subprocess         │
└──────────────────┬──────────────────────────┘
                   │ spawns & IPC (stdio JSON)
┌──────────────────▼──────────────────────────┐
│  Python Subprocess                          │
│  - Audio capture (sounddevice)              │
│  - Whisper transcription (faster-whisper)   │
│  - Streams markdown to ~/Library/App Support│
└─────────────────────────────────────────────┘

Privacy

• All transcription happens locally on your machine
• Audio is never sent to any server
• Transcripts are stored only in ~/Library/Application Support/OhhBrother/
• LLM summarization (if enabled) sends transcript text to Claude or OpenAI

Troubleshooting

"Microphone access denied"

Grant microphone access in System Settings > Privacy & Security > Microphone

Models downloading slowly

First run downloads ~500MB of ML models. This is one-time only.

High CPU usage

Transcription is CPU-intensive. Processing happens in 5-second batches to balance latency and resource usage.

License

MIT