Intent-driven, infinitely adaptive short-form video feed.
MindScroll reimagines the short-video experience. Instead of an opaque algorithm deciding what you watch, you set the intent.
Type or speak exactly what you want—"explain quantum computing simply," "calm lo-fi beats," or "high-intensity workout motivation." MindScroll instantly builds a curated YouTube feed tailored to your prompt. As you watch, it analyzes your engagement signals in real-time, constantly refining and adapting the feed to match your evolving interests.
No doom-scrolling. Just mindful, intent-driven discovery.
- 🎯 Intent-Driven Curation: Describe your mood or learning goal, and let AI build the perfect feed.
- 🎙️ Voice Input (Optional): Seamlessly dictate your intent using NVIDIA Riva gRPC integration.
- ⚡ Real-Time Adaptation: The feed evolves dynamically based on what you watch and like.
- 🧠 AI-Powered Scoring: Google Gemini analyzes video transcripts, metadata, and intent to rank content intelligently.
- 🚀 High Performance: Built on Next.js 16 App Router, styled with Tailwind CSS v4, and animated with Framer Motion.
- 🛡️ Production Ready: Includes rate limiting, Upstash Redis caching, MongoDB Atlas integration, and Vercel Cron jobs.
git clone https://github.com/sayan365/mindscroll.git
cd mindscrollnpm installCopy the example environment file and fill in your keys:
cp .env.example .env.local(See the Environment Variables section below for required keys).
If you don't have a MongoDB Atlas cluster yet, you can quickly spin up a local instance using Docker:
docker compose up -d(MongoDB UI will be available at http://localhost:8081)
npm run devOpen http://localhost:3000 in your browser.
To run MindScroll, you'll need a few API keys. Check out .env.example for detailed instructions.
| Variable | Required | Description |
|---|---|---|
GEMINI_API_KEY |
✅ | Google AI Studio |
YOUTUBE_API_KEY |
✅ | Google Cloud Console (Enable YouTube Data API v3) |
MONGODB_URI |
✅ | MongoDB connection string (Atlas or Local) |
MONGODB_DB |
✅ | Database name (e.g., mindscroll) |
CRON_SECRET |
✅ | Secure random string to protect the Vercel cron endpoint |
UPSTASH_REDIS_REST_URL |
❌ | Upstash Redis URL (Falls back to in-memory if empty) |
UPSTASH_REDIS_REST_TOKEN |
❌ | Upstash Redis Token |
NVIDIA_API_KEY |
❌ | Needed for voice transcription |
NEXT_PUBLIC_HAS_NVIDIA |
❌ | Set to "true" to enable the UI mic button |
MindScroll is optimized for Vercel.
- Push your repository to GitHub.
- Import the project in your Vercel Dashboard.
- Add all the required Environment Variables.
- Make sure to generate a strong
CRON_SECRET(e.g.,openssl rand -hex 32) and add it to your Vercel environment variables. - Deploy!
Note: Vercel Cron will automatically trigger
/api/cron/refresh-poolsonce per day to keep the video pools fresh, using theCRON_SECRETfor authentication.
MindScroll's backend is a symphony of AI and traditional APIs:
- Input: User submits an intent via text or voice.
- Analysis: Google Gemini parses the intent, extracting core topic tags and the desired mood.
- Sourcing: The backend queries the YouTube Data API v3 for relevant videos based on the extracted tags.
- Scoring: Gemini evaluates and scores the retrieved videos against the user's original intent.
- Storage: Sessions, signals, and video metadata are persisted in MongoDB Atlas.
- Delivery: The highly responsive frontend serves the feed, continuously reporting watch/like signals back to the server to refine upcoming content.
We welcome contributions! Please see our Contributing Guidelines for details on how to get started, set up your development environment, and submit pull requests.
This project is licensed under the MIT License - see the LICENSE file for details.