TL;DW (Too Long; Didn't Watch)

Summarize YouTube videos using OpenAI's ChatGPT with intelligent two-phase processing for high-quality, structured summaries.

USAGE:

Usage: tldw [OPTIONS] --video-url <VIDEO_URL>

Options:
  -v, --video-url <VIDEO_URL>
  -e, --engine <ENGINE>        [default: 4]
  -p, --prompt <PROMPT>        [default: 0]
  -h, --help                   Print help
  -V, --version                Print version

Requirements

• An environment variable named OPENAI_API_KEY must be defined, containing an OpenAI API secret key.
• yt-dlp must be installed and available in your system's PATH.

Features

• Two-Phase Processing: Uses Map-Reduce pattern for better context preservation
• Structured Output: Generates both bullet points and narrative summaries
• Cost-Efficient: Smart model selection (GPT-3.5 for extraction, GPT-4 for synthesis)
• Cost Control: Built-in cost estimation and optional spending limits
• Semantic Chunking: Intelligent content splitting that preserves meaning
• Quality Focus: Prioritizes summary quality over real-time streaming

Limitations

• Videos without subtitles cannot be processed.
• The quality of summaries based on auto-generated subtitles may vary.
• Results default to English, even if the input language is different.
• Processing time is longer than simple chunking (but produces much better results).
• Requires sufficient OpenAI API quota for two-phase processing.

Architecture: Two-Phase Processing

%%{ init : { "theme" : "forest", "flowchart" : { "curve" : "stepBefore" }}}%%
graph TD
    A[Get YouTube URL] --> B[Download subtitles via yt-dlp]
    B --> C[Clean and process subtitles]
    C --> D[Estimate cost and check limits]
    D --> E[Phase 1: Extract key information]
    E --> F[Create semantic chunks]
    F --> G[Extract themes, points, quotes using GPT-3.5]
    G --> H[Combine extracted content]
    H --> I[Phase 2: Synthesize summary]
    I --> J[Generate structured output using GPT-4]
    J --> K[Format as markdown with bullets + narrative]
    K --> L[Save structured summary]

Output Format

The tool generates structured summaries with:

# Video Title

## Key Themes
- **Theme 1**: Brief description
- **Theme 2**: Brief description

## Detailed Points
### Theme 1
- Key point 1
- Key point 2

### Theme 2
- Key point 1
- Key point 2

## Narrative Summary
[Coherent narrative connecting all themes with logical flow]

## Key Quotes & Insights
- "Important quote 1"
- "Important quote 2"

Cost Management

• Smart Model Selection: Uses GPT-3.5-turbo for extraction (cheaper) and GPT-4 for synthesis (better quality)
• Cost Estimation: Shows estimated cost before processing
• Spending Limits: Built-in $5 default limit (configurable)
• Linear Scaling: Cost scales linearly with content size, not exponentially