YouTube Notes — Project Skill

Project Context

• •App: Single Streamlit script at app.py. Run: streamlit run app.py from repo root.
• •Flow: Paste URLs → extract video ID (regex for youtube.com/watch?v=ID, youtu.be/ID) → fetch transcript via YouTubeTranscriptApi.get_transcript → summarize with Groq (llama-3.1-8b-instant).
• •Env: GROQ_API_KEY in .env; if unset, app shows transcript only and warns.

When Adding Features

1. •New URL formats: Extend the regex in YOUTUBE_ID_PATTERN or the logic in extract_video_id; keep invalid-URL handling and user-facing "Invalid YouTube URL."
2. •New LLM backend (e.g. Ollama): Add a branch in the summarization path (e.g. check env like USE_OLLAMA or model choice); call Ollama API or local endpoint with the same prompt used for Groq; keep transcript fetch and error handling unchanged.
3. •Export (copy / download): Add a Streamlit control (e.g. st.download_button or copy-to-clipboard) that uses the current notes text; do not change transcript or summarization logic.
4. •Transcript-only mode: Already supported when GROQ_API_KEY is unset; optional: add an explicit "Show transcript only" toggle that skips the Groq call when set.

Conventions to Preserve

• •Transcript: youtube-transcript-api only; no YouTube Data API, no Whisper unless explicitly requested.
• •One summarization prompt tuned for programming notes (bullets, concepts, takeaways).
• •~1.5 s delay between transcript requests for multiple URLs.
• •Errors: "This video has no captions available." for missing transcript; "Invalid YouTube URL." for bad URLs.

Key Files

• •app.py — UI, URL parsing, transcript fetch, Groq summarization
• •AGENTS.md — Layout, conventions, quick reference
• •README.md — Setup and usage for users