BetterTube

Local Python web app that learns your YouTube watch preferences from links you provide, then suggests 1-10 similar videos.

Features

train flow:
- Enter a YouTube link.
- Or upload a .txt file with many YouTube links to bulk train in one shot.
- App opens the video metadata and stores:
  - title
  - creator/channel
  - duration
  - description
  - tags
  - subject signals (categories or inferred keywords)
suggest flow:
- Builds a preference profile from all trained videos.
- Searches YouTube for similar candidates.
- Scores candidates by metadata similarity and returns top 1-10.
- You can like or dislike each suggestion to fine-tune future results.
- Liked suggestions are added to your training log and removed from the active suggestion list.
- Disliked suggestions are removed and used as negative training signals (title, creator, tags, description/captions context).
- Remaining suggestions stay visible and are not auto-filled after you make a decision.
- You can clear recommendation logs (and optionally clear feedback).
Metadata depth:
- Pulls title, creator, duration, tags, description, subjects.
- Also tries chapter titles and closed-caption/auto-caption text to build context_text when descriptions are thin.
Local SQLite database (bettertube.db) for trained videos and recommendation logs.

cd BetterTube
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
python app.py

The app uses yt-dlp to fetch metadata, so internet access is required.
Some videos may block metadata extraction depending on region, age restrictions, or YouTube anti-bot checks.