genai-yt

AI-powered YouTube transcript extractor and analyzer using Claude Code or Cursor CLI.

Features

• Transcript extraction - Extract raw transcripts from YouTube videos
• Speech-to-text fallback - Automatically falls back to yt-dlp + whisper-cpp when subtitles are unavailable
• AI-powered analysis - Summarize, extract insights, translate, or take notes using AI
• Multi-provider support - Claude Code CLI and Cursor CLI
• Custom prompts - Ask any question about a video with ask command
• Multi-language - Specify transcript language and translation target

How It Works

flowchart TD
    A[Start: genai-yt] --> B{Command?}
    B -->|transcript| C[Fetch YouTube Transcript]
    B -->|summary/insights/translate/memo/ask| D[Fetch YouTube Transcript]
    C --> E{Subtitles available?}
    D --> F{Subtitles available?}
    E -->|Yes| G[Output raw text or timestamped]
    E -->|No| H[yt-dlp + whisper-cpp]
    H --> G
    F -->|Yes| I[Build AI Prompt]
    F -->|No| J[yt-dlp + whisper-cpp]
    J --> I
    I --> K{Provider}
    K -->|Claude Code| L[Claude Code CLI]
    K -->|Cursor CLI| M[Cursor CLI]
    L --> N[Display AI Response]
    M --> N

Prerequisites

For AI commands, you need at least one of:

• Claude Code CLI - Anthropic's official CLI
• Cursor Agent CLI - Cursor's agent CLI (command: agent)

Speech-to-Text (optional)

For videos without subtitles, install these tools for automatic fallback:

brew install yt-dlp whisper-cpp ffmpeg

Download a Whisper model (one-time setup):

mkdir -p ~/.local/share/whisper-cpp
cd ~/.local/share/whisper-cpp

# Base model (~142MB, fastest)
curl -LO https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin

# Or large-v3-turbo (~809MB, most accurate)
curl -LO https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-large-v3-turbo.bin

Installation

# Global installation
npm install -g genai-yt

# Or use directly with npx (no installation required)
npx genai-yt transcript "<url>"

Usage

Extract Transcript (No AI)

# Raw transcript text
genai-yt transcript "<url>"

# With timestamps
genai-yt transcript "<url>" --timestamps

# Specify language
genai-yt transcript "<url>" --lang ko

AI-Powered Commands

All AI commands require -p, --provider option.

# Summarize a video
genai-yt summary "<url>" -p claude-code

# Extract key insights
genai-yt insights "<url>" -p cursor-cli

# Translate transcript
genai-yt translate "<url>" -p claude-code --lang en

# Convert to organized notes
genai-yt memo "<url>" -p claude-code

# Ask a custom question
genai-yt ask "<url>" -p claude-code --prompt "Extract all investment advice from this video"

Common Options

# Specify AI model
genai-yt summary "<url>" -p cursor-cli --model claude-4.5-sonnet

# Specify transcript language
genai-yt summary "<url>" -p claude-code --transcript-lang ja

Commands

| Command | Description | AI Required | |---------|-------------|-------------| | transcript <url> | Extract raw transcript | No | | summary <url> | Summarize video content | Yes | | insights <url> | Extract key insights and action items | Yes | | translate <url> | Translate transcript to target language | Yes | | memo <url> | Convert to organized notes | Yes | | ask <url> | Custom prompt-based analysis | Yes |

Options

transcript

| Option | Description | Default | |--------|-------------|---------| | --lang <lang> | Transcript language (e.g., ko, en, ja) | auto | | --timestamps | Include timestamps | false |

AI Commands (summary, insights, translate, memo, ask)

| Option | Description | Default | |--------|-------------|---------| | -p, --provider <provider> | AI provider (claude-code or cursor-cli) | required | | --model <model> | Model to use | provider default | | --transcript-lang <lang> | Transcript language | auto |

translate (additional)

| Option | Description | Default | |--------|-------------|---------| | --lang <lang> | Target language | en |

ask (additional)

| Option | Description | Default | |--------|-------------|---------| | --prompt <prompt> | Custom prompt | required |

Transcript Fallback

When YouTube subtitles are not available, genai-yt automatically falls back to local speech-to-text:

1. Downloads audio using yt-dlp (converted to 16kHz WAV)
2. Transcribes locally using whisper-cpp (no API key required)
3. Returns the transcribed text in the same format

The model is searched in these locations (first match wins):

| Priority | Path | |----------|------| | 1 | WHISPER_MODEL environment variable | | 2 | ~/.local/share/whisper-cpp/ggml-large-v3-turbo.bin | | 3 | ~/.local/share/whisper-cpp/ggml-base.bin | | 4 | ~/.local/share/whisper-cpp/ggml-small.bin | | 5 | ~/.local/share/whisper-cpp/ggml-medium.bin | | 6 | ~/.local/share/whisper-cpp/ggml-large-v3.bin |

Examples

# Get Korean transcript with timestamps
genai-yt transcript "https://youtu.be/dQw4w9WgXcQ" --lang ko --timestamps

# Summarize a tech talk
genai-yt summary "https://www.youtube.com/watch?v=VIDEO_ID" -p claude-code

# Translate Japanese video to English
genai-yt translate "https://youtu.be/VIDEO_ID" -p claude-code --transcript-lang ja --lang en

# Extract investment tips
genai-yt ask "https://youtu.be/VIDEO_ID" -p cursor-cli --prompt "List all actionable investment tips"

# Video without subtitles (auto STT fallback)
genai-yt transcript "https://www.youtube.com/watch?v=VIDEO_ID"

Requirements

• Node.js >= 18.0.0
• Claude Code CLI or Cursor CLI installed and authenticated (for AI commands)
• yt-dlp + whisper-cpp + ffmpeg (optional, for videos without subtitles)

License

MIT