CLI tool that generates YouTube chapters from video transcripts using LLMs. Works with local models via Ollama by default, with optional cloud provider support.

pip install -e .

With OpenAI support:

pip install -e ".[openai]"

With dev/test dependencies:

pip install -e ".[all]"

1. Install and start Ollama: https://ollama.com/
2. Pull a model:

ollama pull llama3.1

3. Generate chapters:

chaptergen generate --input transcript.srt --format youtube

export OPENAI_API_KEY=sk-...
chaptergen generate --input transcript.srt --provider openai --model gpt-4o-mini --format youtube

chaptergen generate --input <file> [options]

Options:

Verify that your provider is reachable and the model is available:

chaptergen check --provider ollama --model llama3.1

00:00 Introduction
02:15 Setting Up the Project
05:30 Writing Core Logic

Ready to paste into a YouTube video description:

Chapters:
00:00 Introduction
02:15 Setting Up the Project
05:30 Writing Core Logic

{
  "provider": "ollama",
  "model": "llama3.1",
  "chapters": [
    {"timestamp": "00:00", "start_seconds": 0, "title": "Introduction"},
    {"timestamp": "02:15", "start_seconds": 135, "title": "Setting Up the Project"}
  ]
}

• llama3.1 — Good balance of speed and quality
• llama3.1:70b — Higher quality, needs more VRAM
• mistral — Fast, decent results

• gpt-5-mini — Cost-effective, good quality
• gpt-5.3 — Best quality

pip install -e ".[all]"
pytest

"Cannot connect to Ollama" — Make sure Ollama is running (ollama serve) and accessible at the expected URL.

"Model not found" — Pull the model first: ollama pull llama3.1

"No valid chapters found" — The LLM failed to return structured output. Try a different model or re-run (local models can be inconsistent). Adding --temperature 0 helps with determinism.

Chapters look wrong — Adjust --min-gap to control spacing, or use --max-chapters to limit count.