meeting-assist

An intelligent meeting assistant for macOS that listens to live meetings and provides real-time insights, suggestions, and conversation tracking.

Quick Start: See QUICKSTART.md for the fastest way to get started.

AI Models: Optimized for speed and cost. See AI_MODEL_GUIDE.md for configuration options.

Features

• Live Meeting Audio Capture: Listens to meetings on macOS in real-time
• Intelligent Talking Points: AI-powered suggestions for relevant talking points during the conversation
• Question Detection: Automatically highlights questions from the conversation
• Urgent Question Flagging: Identifies and flags questions directed specifically at you as urgent
• Full Transcript Recording: Maintains a complete record of the entire conversation
• Action Item Extraction: Proposes todos and action items from the discussion
• Live Summary: Keeps an updated conversation summary with high-level bullet points
• AI Integration: Leverages AI for intelligent summarization and analysis

Tech Stack

• UI Framework: Textual - Python TUI framework for building rich terminal applications
• Platform: macOS
• AI Integration: For summarization and natural language processing

Installation

Prerequisites

• Python 3.10 or higher
• macOS 14.4+ (Sonoma or later) for native system audio capture
  • Older macOS versions will work in mic-only mode
• Anthropic API key

Setup

1. Clone the repository:

git clone <repository-url>
cd meeting-assist

2. Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # On macOS/Linux

3. Install dependencies:

pip install -e .

4. Set up your environment variables:

cp .env.example .env
# Edit .env and add your ANTHROPIC_API_KEY

5. On first run, faster-whisper will download the Whisper model (base model ~150MB)

6. First-time permission: Grant Screen Recording permission when prompted

   • Required for native system audio capture (Core Audio Taps)
   • One-time setup - macOS will prompt automatically
   • Or manually enable: System Settings > Privacy & Security > Screen Recording

Usage

Running the Application

Easiest method - Use the start script:

./start.sh

The start script automatically:

• Creates virtual environment if needed
• Installs dependencies if needed
• Checks for API key configuration
• Loads environment variables from .env
• Runs the application

Try it with mock data (no API key needed):

./start.sh --mock

This populates the UI with sample meeting data so you can explore the interface without setting up API keys or recording audio.

Manual method:

# Activate virtual environment
source venv/bin/activate

# Set your API key
export ANTHROPIC_API_KEY=your_key_here

# Run the application
meeting-assist

Or use the Python module directly:

python -m meeting_assist.main

Controls

• Start Recording: Click "Start Recording" button or press r
• Stop Recording: Click "Stop Recording" button or press r again
• Clear All: Press c to clear all panels
• Quit: Press q

Command Line Options

./start.sh --mock    # Run with mock data (no API key needed)
./start.sh --clear   # Clear stored data before starting
./start.sh --help    # Show all options

Features in Action

Once recording starts:

• Live Transcript: See real-time transcription in the left panel
• Questions: Detected questions appear in the top-right panel
  • Questions directed at you are marked as URGENT with 🚨
• Talking Points: AI-suggested conversation points appear below questions
• Action Items: Extracted todos appear in the action items panel
• Summary: Live-updated conversation summary with key bullet points

Configuration

The application uses sensible defaults optimized for speed and cost:

• Audio: 16kHz sample rate, mono channel
• Transcription: Base Whisper model (good balance, ~16x realtime speed)
  • Free (local processing)
  • Can use "tiny" model for maximum speed if needed
• AI: Claude 3.5 Haiku (fastest + cheapest Claude model)
  • ~$0.25 per million input tokens
  • ~$1.25 per million output tokens
  • Sub-second response times
• Analysis: Every 15 seconds during recording

Model Selection Priority:

1. Speed (fast enough for real-time processing)
2. Cost (affordable for 1+ hour meetings)

To customize, edit the parameters in src/meeting_assist/ui/app.py

Visual Transcript Layout

The transcript shows conversation in a chat-style interface:

• Others' speech: Left-aligned, white text
• Your speech: Right-aligned, cyan text

This requires capturing both:

• Microphone input (your speech) - works immediately
• System audio output (others' speech) - uses native Core Audio Taps (macOS 14.4+)

Dual-Stream Audio Capture

Native System Audio (macOS 14.4+):

• Uses Core Audio Taps - Apple's native API for system audio capture
• Zero third-party dependencies (no BlackHole or Loopback needed!)
• One-time Screen Recording permission grant
• Automatically captures audio from all apps (Zoom, Teams, Chrome, etc.)
• See CORE_AUDIO_TAPS.md for technical details

Toggle Controls:

• 🎤 Microphone: Enable/disable your microphone (captured as YOU)
• 🔊 System Audio: Enable/disable system audio (captured as OTHERS)
• Mix and match: test with mic-only, system-only, or both

Older macOS Versions (< 14.4):

• Falls back to mic-only mode automatically
• All audio appears as "you" (right-aligned, cyan)
• For dual-stream on older macOS, see SYSTEM_AUDIO_SETUP.md (BlackHole setup)

License

See LICENSE file for details.