transcribee – OpenClaw Skill

---

name: transcribee description: Transcribe YouTube videos and local audio/video files with speaker diarization. Use when user asks to transcribe a YouTube URL, podcast, video, or audio file. Outputs clean speaker-labeled transcripts ready for LLM analysis.

Transcribee

Transcribe YouTube videos and local media files with speaker diarization via ElevenLabs.

Usage

# YouTube video
transcribee "https://www.youtube.com/watch?v=..."

# Local video
transcribee ~/path/to/video.mp4

# Local audio
transcribee ~/path/to/podcast.mp3

Always quote URLs containing & or special characters.

Output

Transcripts save to: ~/Documents/transcripts/{category}/{title}-{date}/

File	Use
transcription.txt	Speaker-labeled transcript
transcription-raw.txt	Plain text, no speakers
transcription-raw.json	Word-level timings
metadata.json	Video info, language, category

Supported Formats

• Audio: mp3, m4a, wav, ogg, flac
• Video: mp4, mkv, webm, mov, avi
• URLs: youtube.com, youtu.be

Dependencies

brew install yt-dlp ffmpeg

Troubleshooting

Error	Fix
yt-dlp not found	brew install yt-dlp
ffmpeg not found	brew install ffmpeg
API errors	Check .env file in transcribee directory

transcribee 🐝

Open source macOS transcriber for YouTube, Instagram Reels, TikTok, and local media — evolves a self-organizing knowledge base.

transcribee "https://youtube.com/watch?v=..."
transcribee "https://instagram.com/reel/..."
transcribee "https://vt.tiktok.com/..."
transcribee ~/Downloads/podcast.mp3

Over time, your ~/Documents/transcripts/ folder naturally evolves into a personal library:

transcripts/
├── AI-Research/
│   ├── ilya-sutskever-agi-2024/
│   └── anthropic-constitutional-ai/
├── Startups/
│   ├── ycombinator-how-to-get-users/
│   └── pmarca-founder-mode/
└── Health/
    └── huberman-sleep-optimization/

Each transcript is speaker-labeled and ready to paste into ChatGPT, Claude, or any LLM.

Why 🍯

I consume a lot of video content — YouTube, Instagram, TikTok, podcasts, interviews. I wanted to:

• Ask questions about videos in LLMs
• Have all that knowledge searchable and organized
• Not do any manual work to maintain it

transcribee does exactly that. Transcribe once, knowledge stays forever.

Features 🪻

• Transcribes YouTube, Instagram Reels, TikTok, and local audio/video files
• Speaker diarization — identifies different speakers
• Auto-categorizes transcripts using Claude based on content
• Builds a knowledge library that organizes itself over time

Use with Clawdbot 🤖

transcribee is available as a Clawdbot skill. Just ask your agent to transcribe any YouTube video:

"Transcribe this video: https://youtube.com/watch?v=..."

Install the skill

# Install from ClawdHub (recommended)
clawdhub install transcribee

# Or clone manually
git clone https://github.com/itsfabioroma/transcribee.git ~/.clawdbot/skills/transcribee

Make sure you have the dependencies installed (brew install yt-dlp ffmpeg) and API keys configured.

Quick Start 🪺

# Install dependencies (macOS)
brew install yt-dlp ffmpeg
pnpm install

# Configure API keys
cp .env.example .env
# Add your ElevenLabs + Anthropic API keys to .env

# Transcribe anything
transcribee "https://youtube.com/watch?v=..."
transcribee "https://instagram.com/reel/..."
transcribee "https://vt.tiktok.com/..."
transcribee ~/Downloads/podcast.mp3
transcribee ~/Videos/interview.mp4

Shell alias (recommended)

Add to ~/.zshrc:

alias transcribee="noglob /path/to/transcribee/transcribe.sh"

Output 🍯

Each transcript saves to ~/Documents/transcripts/{category}/{title}/:

File	What it's for
transcript.txt	Speaker-labeled transcript — paste this into your LLM
metadata.json	Video info, language, auto-detected theme

Raw JSON (optional)

For power users who need word-level timestamps and confidence scores:

transcribee --raw "https://youtube.com/watch?v=..."

This adds transcript-raw.json with the full ElevenLabs response.

How it works 🐝

1. Downloads audio from YouTube (yt-dlp) or extracts from local video (ffmpeg)
2. Transcribes with ElevenLabs (scribe_v1_experimental with speaker diarization)
3. Claude analyzes content and existing library structure
4. Auto-categorizes into the right folder
5. Saves transcript files with metadata

Requirements

• macOS (tested on Sonoma)
• Node.js 18+
• yt-dlp — brew install yt-dlp
• ffmpeg — brew install ffmpeg
• ElevenLabs API key — for transcription
• Anthropic API key — for auto-categorization

Supported formats

Type	Formats
Audio	mp3, m4a, wav, ogg, flac
Video	mp4, mkv, webm, mov, avi
URLs	youtube.com, youtu.be, instagram.com/reel, tiktok.com

---

bzz bzz 🐝