MeetMind

An AI meeting assistant that joins your video calls as a participant, records and transcribes the conversation, writes a summary, and answers questions about the meeting after it ends.

MeetMind is built as a full-stack Next.js application with a typed tRPC API, PostgreSQL persistence (Neon + Drizzle), Stream-powered video and chat, OpenAI-powered intelligence, and Inngest background processing.

Features

• Custom AI agents — create agents with names and behavioral instructions.
• AI-assisted live meetings — the agent joins Stream Video calls in real time via the OpenAI Realtime API.
• Automatic capture — transcription, closed captions, and recording are enabled on every call.
• Background summaries — an Inngest pipeline parses the transcript, attributes speakers, and generates a Markdown summary with GPT-4o.
• Post-meeting chat — keep asking the agent questions about the meeting through Stream Chat.
• Dashboard — manage agents and meeting history with search, filters, and pagination.
• Auth — email/password plus GitHub and Google OAuth via Better Auth.

Quick Start

Prerequisites

• Node.js 20+
• A PostgreSQL database (Neon recommended — the project uses the Neon HTTP Drizzle driver)
• A Stream account with Video and Chat enabled
• An OpenAI API key
• Optional: GitHub/Google OAuth apps, ngrok (for local webhooks)

Installation

git clone <repository-url>
cd meetmind
npm install

Create .env.local in the project root using the Configuration table below, then push the schema and start the dev server:

npm run db:push
npm run dev

The app runs at http://localhost:3000.

Stream Webhooks (local development)

Stream cannot reach localhost, so expose the dev server with a tunnel:

ngrok http 3000          # or: npm run dev:webhook (uses the reserved domain in package.json)

Set the Stream webhook URL to https://<your-ngrok-domain>/api/webhook and subscribe to at least: call.session_started, call.session_participant_left, call.session_ended, call.transcription_ready, call.recording_ready, and message.new.

To run Inngest functions locally, start the dev server in a separate terminal:

npx inngest-cli dev

Configuration

Variable	Required	Used By	Description
DATABASE_URL	Yes	Drizzle, Better Auth	PostgreSQL connection string.
BETTER_AUTH_SECRET	Yes	Better Auth	Secret used by Better Auth. Use a long random value.
BETTER_AUTH_URL	Yes	Better Auth	Base URL of the app. Use http://localhost:3000 locally.
NEXT_PUBLIC_APP_URL	Yes	tRPC	Public app URL used by the server-side tRPC client URL builder.
GITHUB_CLIENT_ID	Optional	Better Auth	GitHub OAuth client ID.
GITHUB_CLIENT_SECRET	Optional	Better Auth	GitHub OAuth client secret.
GOOGLE_CLIENT_ID	Optional	Better Auth	Google OAuth client ID.
GOOGLE_CLIENT_SECRET	Optional	Better Auth	Google OAuth client secret.
NEXT_PUBLIC_STREAM_VIDEO_API_KEY	Yes	Stream Video client/server	Public Stream Video API key.
STREAM_VIDEO_SECRET_KEY	Yes	Stream Video server	Secret key used by server-side Stream Video APIs and webhook verification.
NEXT_PUBLIC_STREAM_CHAT_API_KEY	Yes	Stream Chat client/server	Public Stream Chat API key.
STREAM_CHAT_SECRET_KEY	Yes	Stream Chat server	Secret key used by server-side Stream Chat APIs.
OPENAI_API_KEY	Yes	OpenAI, Stream OpenAI integration, Inngest	API key for live AI, summary generation, and chat responses.

Scripts

Command	Description
npm run dev	Starts the Next.js development server.
npm run build	Builds the app for production.
npm run start	Starts the production server after a build.
npm run lint	Runs ESLint.
npm run lint:fix	Runs ESLint and applies automatic fixes.
npm run format	Formats the repository with Prettier.
npm run format:check	Checks formatting without writing changes.
npm run db:push	Pushes the Drizzle schema to the database.
npm run db:studio	Opens Drizzle Studio.
npm run dev:webhook	Starts an ngrok tunnel using the reserved domain in package.json.
npm run prepare	Sets up Husky hooks.

There is currently no test script defined in package.json.

Documentation

Document	What it covers
ARCHITECTURE.md	C4 diagrams, data-flow sequences, infrastructure, engineering challenges, and known limitations.
CONTEXT.md	Dense primer for contributors and AI assistants: stack versions, codebase map, patterns, invariants, and glossary.
MeetMind-Project-Overview.md	Product requirements, feature breakdown, detailed workflows, database design, and tech-stack rationale.
DESIGN.md	Design philosophy, semantic CSS tokens, and visual system.
docs/how-tos/how-to-deploy.md	Production deployment steps and checklist.
docs/runbooks/	Troubleshooting: webhooks/tunnels, transcript pipeline stalls, call/AI failures, and local setup & configuration.
docs/concepts/	Deep-dives: durable execution, realtime voice agents, and speaker diarization.
docs/adr/	Architecture Decision Records.

License

This is currently a private project.