Discover high-signal YouTube videos for any learning goal, then chat with a local LLM grounded on the video transcript.

✨ Features

• Query refinement
  Turn a short learning goal into a sharp YouTube search query using your local Ollama model.
• YouTube search
  Fetch relevant videos via the YouTube Data API v3.
• Transcript fetch
  Prefers manual transcripts, then auto-generated, then translated-to-English.
• Grounded chat
  Ask questions about the selected video; answers are based on its transcript.
• Simple, responsive UI
  Built with Streamlit.

🚀 Quick Start

1. Prerequisites

• Python 3.11+
• An Ollama server running locally or reachable over HTTP
• A YouTube Data API v3 key

2. Clone & install

git clone https://github.com/Anishrkhadka/asktube.git
cd asktube

3. Configure environment

Create a .env file in the project root:

# Required: YouTube Data API key
YOUTUBE_API_KEY=YOUR_API_KEY_HERE

# Optional: Ollama host (defaults to http://localhost:11434)
OLLAMA_HOST=http://localhost:11434

4. Ensure a model is available in Ollama

The app defaults to gemma3:12b (falls back to discovered tags). Pull at least one model:

ollama pull gemma3:12b
# or another compatible model, e.g. llama3.1:8b, mistral:7b

🐳 Docker

Build:

docker compose up --build 

🎯 Usage

1. In the text area, describe what you want to learn.
2. Choose number of videos, model, sort order, and duration filter.
3. Click Find videos to fetch results.
4. Pick a video from the sidebar. Transcript loads automatically (if available).
5. Use the chat box to ask questions grounded on the transcript.

🔑 Environment Notes

• YouTube API key → Create in Google Cloud Console (enable YouTube Data API v3, create API key).
• Ollama → Install Ollama, run ollama serve, and pull a model with ollama pull <model>.

📄 License

MIT License © 2025 AskTube