Google Meet Meeting Bot

This is a PoC that joins a Google Meet, scrapes live captions, sends captions to OpenAI for summarization, and stores the transcript and summary in PostgreSQL. I've added a small frontend to allow users to submit meeting links.

If you're interested in the process, reasoning, demos, and more, check out the blog.

Hosted Google Meet Bot

If you don't want to host a bot yourself, check out Recall.ai

Tech Stack

• Node.js / TypeScript
• Playwright (headless browser bot)
• OpenAI API
• PostgreSQL
• Docker + Docker Compose
• Prisma ORM
• Express (API) + simple HTML (frontend)

Steps

• Join Meet using Playwright
• Scrape captions from the DOM
• Flush transcript to PostgreSQL
• Call OpenAI for summary generation
• Store summary in PostgreSQL
• Create web UI so that meeting links can be submitted

How to run the project

1. Install prereqs

   • Docker ()

   • Install node and a package manager

     • Option 1:
       • Brew install node by opening terminal and running brew install node then confirming installation by running node -v and npm -v (if you don't have homebrew installed, start by installing homebrew on your machine)
     • Option 2:
       • Open terminal and run sudo apt install nodejs then run node -v to confirm installation
       • npm on Linux: run sudo apt install npm then npm -v to confirm installation

   • Git

   • A second Google account to join meetings (you will manually start a meeting using one account of your choosing and then your bot will need an account to join the meetings from)

   • An OpenAI API key

   • Install Playwright npm install -D playwright then npx playwright install

2. Clone the Repository

git clone https://github.com/recallai/google-meet-meeting-bot.git
cd google-meet-meeting-bot

3. run npm install

4. To copy the .env.sample file and rename to .env in root by running this:

   cp .env.sample .env
   

   replacing the placeholder values for your own values:

   DATABASE_URL=postgresql://meetingbot:supersecret@postgres:5432/meetingbotpoc
   OPENAI_API_KEY=your-openai-api-key
   GOOGLE_ACCOUNT_USER=your-second-google-email
   GOOGLE_ACCOUNT_PASSWORD=your-second-google-password
   

5. Run the generate-auth.js script in scripts/

npm run gen:auth

Do NOT commit your auth.json or .env file to Git. I've already added both to .gitignore

7. Run your code:

docker-compose build --no-cache
docker compose up -d

If you get an error when you run the above, make sure your Docker desktop app is open (the app you installed at the beginning)

8. Run Database Migrations

Prisma's migration files are already included in the repo. To apply copy and paste the following in terminal (the first cmd opens a shell and the second deploys the migration):

docker compose exec backend sh
npx prisma migrate deploy

This will apply the schema to your local PostgreSQL instance (spun up by Docker). To confirm you can try the cmds in the appendix

and you should see tables now, but that is not necessary.

Note: If you're modifying the schema yourself, use npx prisma migrate dev instead to generate new migrations.

9. Exit out of your shell by typing exit and hitting enter then re-run your code:

docker-compose build --no-cache
docker compose up -d

10. Open a second terminal window and run

cd src/frontend
npm install
npm run dev

11. Start a Google Meet

• Start a meeting with your primary Google account (not the bot account you created)
• copy the url before the '?' (put in a note or somewhere you can return to)
• Go to the "Host Controls" in the bottom right-hand corner
• Select "Open" in "Meeting Access"

12. Navigate to your basic frontend

• Open a new tab
• Paste the following url: http://localhost:5173
• Copy the meeting url you stored in the previous step
• Paste it into your bar and hit submit

13. Conduct your meeting

• Make sure you are unmuted in the Google Meet tab you have open
• Have a conversation and when you want your bot to leave, either end the meeting or say "Notetaker, please leave"
• The bot will send the transcript to OpenAI if you've provided a valid API key and your summary will be stored.

14. Checking your data to access postgres go to the terminal window where you built docker and run:

docker exec -it meetingbot-db psql -U meetingbot -d meetingbotpoc

then

• To see your meeting summary after the call:

SELECT "meetingId",
       "generatedAt",
       "model",
       "summaryText"
FROM   "MeetingSummary"
ORDER  BY "generatedAt" DESC
LIMIT  1;

• To see your transcript:

SELECT t."meetingId",
       t."createdAt",
       json_agg(
         json_build_object(
           'start',   s.start,
           'end',     s."end",
           'speaker', s.speaker,
           'text',    s.text
         )
         ORDER BY s.start
       ) AS segments
FROM   "MeetingTranscript" t
JOIN   "Segment"           s USING ("meetingId")
WHERE  t."meetingId" = (
          SELECT "meetingId"
          FROM   "MeetingTranscript"
          ORDER  BY "createdAt" DESC
          LIMIT  1
      )
GROUP  BY t."meetingId", t."createdAt";

REMEMBER to hit 'q' everytime you want to exit out of the specific query results from the queries above and when you are done looking into what is in your database (postgreSQL) type exit again

Happy meeting!

Project Structure

google-meet-meeting-bot/
├── scripts/
│   ├── generate-auth.js
├── src/
│   ├── backend/
│   │   ├── migrations/
│   │   ├── server.ts
│   │   ├── package.json
│   │   ├── package-lock.json
│   │   ├── tsconfig.json
│   │   ├── launchBot.ts
│   │   └── schema.prisma
│   ├── bot/
│   │   ├── index.ts
│   │   ├── package.json
│   │   ├── package-lock.json
│   │   └── tsconfig.json
│   └── frontend/
│       ├── index.html   # form
│       ├── main.ts
│       ├── style.css
│       ├── tsconfig.json
│       ├── package.json
│       └── package-lock.json
├── playwright/
│   ├── Dockerfile
│   ├── package.json
│   ├── package-lock.json
│   ├── tsconfig.json
│   ├── runBot.ts
│   ├── storage.ts
│   ├── models.ts
│   └── summarize.ts
├── package.json
├── tsconfig.json
├── .env
├── Dockerfile.be
├── Dockerfile.bot
├── docker-compose.yml
├── auth.json            # created by logging into the site via playwright and then storing credentials
└── README.md

Bonus: Want Something More Scalable?

If you're set making this production-ready or integrating with other platforms (Zoom, Teams, Meet), check out Recall.ai. We provide a Desktop Recorder SDK and multi-platform meeting bot infrastructure, which can simplify and scale what this PoC demonstrates.

I know I'm biased, but I high recommended looking into the Recall.ai API if you're looking to move beyond prototypes or checking out some of our customers and case studies if you're wondering how you might leverage conversation data in your product.

Appendix

To confirm your prisma tables you can run

docker exec -it meetingbot-db psql -U meetingbot -d meetingbotpoc

then

\dt

Huge Thanks To...

Amanda for giving me the opportunity to work on this project. YK for showing me the ropes. Antonio for the eng side onboarding. An an extra huge shoutout to Gerry for his invaluable feedback which has made this so simple to run that my mom (not an engineer) could do it. The entire Recall.ai team for being such a stellar and generous team!