Google Meet Integration

Claire can listen to Google Meet calls and automatically execute coding tasks mentioned during meetings. Say something like “let’s add a new API endpoint for user profiles” and Claire will detect it, extract the task, and execute it.

Two Methods

A Puppeteer bot joins the meeting as a participant and captures captions.

Best for: Org meetings where you need a dedicated bot account.

Setup:

  1. Get IT to create a Google Workspace account (e.g., claire-bot@yourcompany.com)
  2. Run the auth script to save the login session: 
    bun run meet-auth
  3. Log into the bot account in the browser that opens
  4. Navigate to meet.google.com and press Enter
  5. Session saved to meet-auth.json

Usage:

In Slack:

/claire bind-meet https://meet.google.com/abc-defg-hij

The bot joins the meeting and starts listening. Results post to a thread in the channel where you ran the command.

Other commands:

/claire unbind-meet https://meet.google.com/abc-defg-hij  # Disconnect bot
/claire meet-status                                        # Show active bots

Method 2: Firefox Extension

A browser extension captures captions from meetings you’re personally in.

Best for: When you don’t have a bot account, or want invisible capture.

Setup:

  1. Open Firefox → about:debugging#/runtime/this-firefox
  2. Click “Load Temporary Add-on…”
  3. Select firefox-extension/manifest.json

Configuration:

Click the extension icon to configure:

Field Description Example
API URL Where Claire is running http://localhost:3000 (local) or http://your-server:3000 (remote)
Slack Channel ID Channel for results. In Slack app: right-click channel → open details → scroll to bottom. Or copy link and extract the C... part C0123456789
Thread Timestamp Optional. Post to specific thread instead of channel 1234567890.123456

Usage:

  1. Join a Google Meet call
  2. Enable captions (Shift+C)
  3. Click extension icon → toggle “Enable Capture”

How Trigger Detection Works

Claire uses two-stage detection:

  1. Keyword matching (fast) - Looks for phrases like “let’s add”, “we should create”, “can we fix”
  2. LLM distillation (accurate) - Extracts the actual task from natural speech

This means you don’t need to say “Claire” - just speak naturally about what you want to build.

Examples that trigger:

  • “Let’s try adding a dark mode toggle”
  • “We should create an endpoint for user settings”
  • “Can we fix that authentication bug?”
  • “I’ll write a function to validate emails”

Examples that don’t trigger:

  • “Yeah that sounds good”
  • “I think we discussed this yesterday”
  • “Let me share my screen”

Configuration

Environment variables in .env:

# Bot display name in meetings
MEET_BOT_DISPLAY_NAME=Claire Bot

# Run browser in headless mode (set false for debugging)
MEET_BOT_HEADLESS=true

# Minimum confidence to execute (0.0 - 1.0)
MEET_MIN_TRIGGER_CONFIDENCE=0.6

# Cooldown between similar triggers (ms)
MEET_COOLDOWN_MS=30000

# HTTP API port for extension
HTTP_PORT=3000

Troubleshooting

Bot can’t join org meetings:

  • Your org likely blocks external guests
  • Solution: Use a Google Workspace account from your org (Method 1 with auth)

Extension not sending captions:

  • Make sure captions are enabled in Meet (Shift+C)
  • Check the API URL is correct and Claire is running
  • Open browser console for error messages

Tasks not being detected:

  • Speak more explicitly about the task (“let’s add X” rather than “maybe we could look at X”)
  • Check MEET_MIN_TRIGGER_CONFIDENCE - lower it if tasks aren’t being picked up

Claire - Apache 2.0