YouTube Digest

Automatically summarize YouTube videos and deliver them as email digests.

Subscribe to your favorite YouTube channels, and get daily email summaries powered by any OpenAI-compatible LLM (OpenAI, local models via Ollama/vLLM/llama.cpp, Anthropic, etc.). Read video content in 5 minutes instead of watching for 20. Self-hosted, privacy-friendly, and fully configurable.

Quick Start

git clone https://github.com/apond308/youtube-digest.git
cd youtube-digest
./install.sh

The installer walks you through everything interactively:

1. Creates a Python virtualenv and installs dependencies
2. Prompts for your API keys (OpenAI-compatible LLM, Gmail)
3. Sets up your channel list and subscriber config
4. Optionally installs as a systemd service and/or Cloudflare tunnel

After setup, run the server:

source .venv/bin/activate
youtube-digest serve

Or run a one-off batch digest:

source .venv/bin/activate
youtube-digest run

How It Works

Daily digest flow:

1. Fetches RSS feeds from all configured YouTube channels
2. For each subscriber, filters to their channels and skips already-sent videos
3. Fetches the transcript, sends it to an LLM for summarization
4. Archives the markdown summary locally and emails the HTML digest
5. Tracks deliveries in SQLite to prevent duplicates

On-demand flow:

1. Open the web UI at http://localhost:4000
2. Paste any YouTube URL and pick a subscriber
3. Summary is generated in the background and emailed when ready

Configuration

All user configuration lives in three files (created by the installer):

File	Purpose
.env	API keys and credentials
channels.yaml	YouTube channels to follow
subscribers.yaml	Email subscribers and their channel preferences

See .env.example, channels.example.yaml, and subscribers.example.yaml for annotated templates.

Environment Variables

Variable	Description	Default
OPENAI_API_KEY	API key for your LLM provider	(required)
OPENAI_BASE_URL	OpenAI-compatible API base URL	https://api.openai.com/v1
MODEL_NAME	Model to use for summarization	gpt-4o-mini
GMAIL_ADDRESS	Gmail address for sending digests	(required)
GMAIL_APP_PASSWORD	Gmail app password (create one here)	(required)
MAX_VIDEOS_PER_DAY	Global max videos per subscriber per run	0 (no limit)

Any OpenAI-compatible API works -- OpenAI, local LLMs via llama.cpp/vLLM/Ollama, Anthropic via a proxy, etc.

CLI Reference

youtube-digest run                # run one digest cycle
youtube-digest serve              # start the web server + daily scheduler
youtube-digest serve --port 9000  # custom port
youtube-digest serve --reload     # auto-reload for development

In server mode, APScheduler runs the daily digest automatically at 05:00 America/Los_Angeles.

API Endpoints

Method	Path	Description
GET	/	Web form for on-demand video submission
POST	/api/summarize	Queue a video: {"url": "...", "subscriber_email": "..."}
GET	/api/health	Health check: {"status": "ok"}

Deployment

For production deployment options (systemd service, Cloudflare tunnel, reverse proxy), see docs/DEPLOYMENT.md.

Project Structure

youtube-digest/
├── install.sh                 # interactive installer
├── .env.example               # env var template
├── channels.example.yaml      # channel config template
├── subscribers.example.yaml   # subscriber config template
├── pyproject.toml
├── deploy/templates/          # systemd + cloudflare templates
├── docs/DEPLOYMENT.md
├── templates/                 # Jinja2 email + web templates
│   ├── email.html
│   └── form.html
└── youtube_digest/            # Python package
    ├── __main__.py            # CLI entry point
    ├── config.py              # settings + channel loading
    ├── models.py              # data models
    ├── pipeline.py            # orchestration logic
    ├── server.py              # FastAPI app + scheduler
    ├── services/
    │   ├── feed.py            # RSS feed fetching
    │   ├── transcript.py      # YouTube transcript retrieval
    │   └── summarizer.py      # LLM summarization
    ├── delivery/
    │   └── email.py           # Gmail SMTP delivery
    └── storage/
        ├── archive.py         # markdown archive of summaries
        ├── database.py        # SQLite sent-video tracking
        └── subscribers.py     # subscriber YAML loading

Dependencies

Package	Purpose
openai	OpenAI-compatible LLM client
youtube-transcript-api	Transcript retrieval
feedparser	RSS/Atom feed parsing
requests	HTTP requests
jinja2	HTML template rendering
markdown	Markdown to HTML conversion
python-dotenv	.env file loading
pyyaml	YAML config parsing
fastapi	Web server framework
uvicorn	ASGI server
apscheduler	Scheduled daily digest

License

MIT