GymBro — YouTube Transcript Summarizer

Extract, timestamp, and organize transcripts from YouTube videos and playlists. Powered by LLMs for intelligent summarization.

Features

• YouTube Playlists — Process entire playlists in one command; each video gets its own transcript file.
• Timestamps — Include timestamps in transcripts for easy reference (--timestamps).
• Organized Output — Creates a folder per playlist/video with individual .txt files named after each video title.
• Combined Transcripts — Generates an all_transcripts.txt merging every video in a playlist.
• Multiple Interfaces — CLI, interactive terminal UI, Python UI, Electron GUI, and Chrome/Firefox extension.
• LLM Summarization — Summarize with OpenAI, Gemini, Claude, Groq, local models, or OpenRouter free tier.
• Media-Aware — Auto-detects video/audio vs page content; supports podcasts, HLS, local files.
• Slide Extraction — Screenshot + OCR + timestamped cards for video sources.

Quick Start

Prerequisites

• Node.js 24+
• yt-dlp (required for YouTube playlist/video extraction)
• ffmpeg (required for slides and media transcription)

# Install yt-dlp and ffmpeg (Windows — winget)
winget install yt-dlp ffmpeg

# macOS (Homebrew)
brew install yt-dlp ffmpeg

Install

# Clone the repo
git clone https://github.com/appo02/GymBro.git
cd GymBro

# Install dependencies
npm install
# or
pnpm install

# Build
pnpm build

Usage

Extract a single video with timestamps

pnpm summarize "https://youtu.be/dQw4w9WgXcQ" --extract --timestamps

Extract an entire playlist

pnpm extract:playlist "https://www.youtube.com/playlist?list=PLxxxxxx"

This creates:

outputs/
└── Playlist Title/
    ├── Video 1 Title.txt
    ├── Video 2 Title.txt
    ├── Video 3 Title.txt
    └── all_transcripts.txt      ← combined file

Interactive UI (no args needed)

# Node.js terminal UI
pnpm extract:ui-simple

# Python UI (no Node prompts dependency needed)
python scripts/playlist_ui.py

# Windows batch shortcut
scripts\playlist_ui.bat

# Electron GUI
pnpm start:gui

Summarize a URL

pnpm summarize "https://example.com" --model google/gemini-2.5-flash

Summarize with length control

pnpm summarize "https://example.com" --length long
pnpm summarize "https://example.com" --length 20k

Supported Inputs

Input	Example
YouTube video	https://youtu.be/dQw4w9WgXcQ
YouTube playlist	https://www.youtube.com/playlist?list=...
Web page	https://example.com/article
Local file	/path/to/file.pdf
Audio/Video	/path/to/audio.mp3
Podcast RSS	https://feeds.npr.org/500005/podcast.xml
HLS stream	https://example.com/master.m3u8
Stdin	echo "text" | pnpm summarize -

Chrome/Firefox Extension

One-click summarizer for the current tab with a Side Panel/Sidebar.

1. Install the CLI globally: npm i -g @steipete/summarize
2. Install the extension from the Chrome Web Store
3. Open Side Panel → copy token → run: summarize daemon install --token <TOKEN>

Configuration

Set your LLM API key:

# Environment variable (pick one)
export OPENAI_API_KEY=sk-...
export GEMINI_API_KEY=...
export GROQ_API_KEY=...
export OPENROUTER_API_KEY=...

Or use the free tier:

summarize refresh-free   # needs OPENROUTER_API_KEY
summarize "https://example.com" --model free

Project Structure

scripts/
├── playlist-extract.js    # Core playlist extraction logic
├── playlist-ui-simple.js  # Interactive terminal UI (readline)
├── playlist-ui.js         # Prompts-based terminal UI
├── playlist_ui.py         # Python alternative UI
└── playlist_ui.bat        # Windows batch launcher

apps/
├── gui/                   # Electron desktop GUI
└── chrome-extension/      # Browser extension (Chrome + Firefox)

src/                       # CLI + library source (TypeScript)
packages/core/             # Shared core library
outputs/                   # Generated transcript files

Scripts

Command	Description
pnpm build	Build the project
pnpm summarize <url>	Summarize a URL (dev mode)
pnpm extract:playlist <url>	Extract playlist transcripts
pnpm extract:ui-simple	Interactive terminal extractor
pnpm start:gui	Launch Electron GUI
pnpm test	Run tests
pnpm check	Format + lint + typecheck + test

How It Works

1. Playlist detection — yt-dlp fetches playlist metadata (video titles, IDs, URLs).
2. Per-video extraction — Each video is processed through the summarize CLI with --extract --timestamps.
3. File organization — Outputs are saved to outputs/<playlist-name>/<video-title>.txt.
4. Combined file — All transcripts are concatenated into all_transcripts.txt with headers.

License

MIT