This CLI tool allows you to query the visual content in any Youtube video through natural language. Providing a Youtube video URL allows the tool to download the video to your local directory, split it a frame per second, and index the frames into a locally spun-up Vector DB which can be queried in natural language.
- First install FFMPEG locally to be able to run the application, i.e. on Mac:
brew install ffmpeg
- Install the requirements within the working directory:
pip install -r requirements.txt
- Generate an OpenAI API key here, and create a .env file with the key, similar to how it is done in
.env.example - Run the terminal app, i.e.
python main.pyOR run it in agent mode usingpython agent.py. In agent mode the agent will autonomously call upon tools for searching through Youtube videos, indexing and analysing them, as well as reviewing already downloaded videos and various metadata.
The program creates a chroma folder in the working directory, where the Chroma Vector DB gets persisted.