Installation

System packages

Install FFmpeg:

sudo apt install ffmpeg

Pull the recommended local model:

ollama pull mistral-small

For faster iteration, pull a smaller model too:

ollama pull qwen2.5:7b

Python dependencies

poetry install

Verify the environment

make check-env

This checks CUDA availability through CTranslate2, lists local Ollama models, and prints the FFmpeg version.

Tip

Face detection uses YuNet through OpenCV and runs on CPU. CUDA matters most for the faster-whisper transcription step.