AI Meet Agent

AI meeting twin — bidirectional realtime audio conversation through Google Meet. Someone speaks, the AI twin hears and responds naturally.

Prerequisites

• Node.js >= 22
• Linux (native or WSL2)
• Google Gemini API key (GEMINI_API_KEY)
• ffmpeg (sudo apt install ffmpeg)
• PipeWire/PulseAudio (typically pre-installed on modern Linux)

Native Linux Only

• v4l2loopback kernel module (for virtual camera)

WSL2 Only

• OBS Virtual Camera (Windows side)
• VB-Audio Virtual Cable (Windows side)

Installation

npm install -g ai-meet-agent

Or run without installing:

npx ai-meet-agent start

Quick Start

1. Set your API key:

   export GEMINI_API_KEY=your-key-here

2. (Optional) Copy and edit the example config:

   cp node_modules/ai-meet-agent/config.example.json ./config.json

3. Start a session:

   ai-meet start

4. Open Google Meet, select "AI Meet Agent Camera" and "AI Meet Agent Mic" as your devices.

Commands

| Command | Description | |---------|-------------| | ai-meet start | Launch a meeting session | | ai-meet start --config <path> | Use a specific config file | | ai-meet start --notes <path> | Load meeting context from markdown | | ai-meet start --role <path> | Load persona from a role file | | ai-meet start --verbose | Enable verbose audio level logging | | ai-meet list-devices | Show available audio/video devices | | ai-meet test-audio | Verify audio device setup | | ai-meet --version | Show version | | ai-meet --help | Show help |

Configuration

All fields are optional — sensible defaults are used when omitted. See config.example.json for the full schema.

Key Fields

| Field | Default | Description | |-------|---------|-------------| | persona.name | "AI Assistant" | Name used in transcripts | | persona.role | "Meeting Participant" | Role context for AI | | persona.instructions | "" | Custom behavior instructions | | ai.model | gemini-2.5-flash-native-audio-latest | Gemini model ID | | devices.camera.videoNr | 10 | v4l2loopback device number |

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | GEMINI_API_KEY | Yes | Google Gemini API key |

.env files in the working directory are loaded automatically.

Error Codes

| Code | Meaning | Common Fix | |------|---------|------------| | 2 | Config error | Check config.json fields — error names the bad field | | 3 | Missing dependency | Install the dependency shown in the error hint | | 4 | AI session error | Check GEMINI_API_KEY and network connection | | 5 | Audio pipeline error | Check PulseAudio/PipeWire setup |

Troubleshooting

"ffmpeg: command not found"

sudo apt install ffmpeg

"v4l2loopback kernel module not loaded"

sudo modprobe v4l2loopback video_nr=10 card_label="AI Meet Agent Camera" exclusive_caps=1 max_buffers=2

"GEMINI_API_KEY not set"

export GEMINI_API_KEY=your-key
# Or create a .env file:
echo "GEMINI_API_KEY=your-key" > .env

WSL2 audio not working

See docs/wsl2-setup.md for Windows-side VB-Cable and OBS configuration.

License

MIT