youtube-to-knowledge

Turn any YouTube video into a transcript, summary, and knowledge graph — powered by Claude Code.

Paste a link. Get structured knowledge.

How it works

1. Fetches the transcript (YouTube API or local Whisper)
2. Generates a structured summary in the video's language
3. Extracts entities and relationships → builds an interactive knowledge graph

Requirements

• Python 3.10+
• Claude Code
• ffmpeg (required for Whisper mode)

Install ffmpeg

OS	Command
macOS	brew install ffmpeg
Ubuntu / Debian	sudo apt install ffmpeg
Fedora	sudo dnf install ffmpeg
Windows	winget install ffmpeg

Setup

git clone https://github.com/velmighty/youtube-to-knowledge
cd youtube-to-knowledge
pip install -r requirements.txt

Open the folder in Claude Code.

Usage

/process https://www.youtube.com/watch?v=VIDEO_ID

Process multiple videos at once:

/process https://youtube.com/watch?v=abc https://youtube.com/watch?v=def

Process an entire playlist:

/process https://www.youtube.com/playlist?list=PLxxx

Already-processed videos are skipped automatically.

Additional commands:

• /video_specialist — deep-dive questions about what was said in a processed video
• /kg_navigator — explore entity connections across videos

Output

Files are saved to vault/content/<channel_name>/:

raw/
  transcript_<video_id>.txt    raw transcript
  metadata_<video_id>.json     title, channel, video ID, language
summary_<video_id>.md          structured summary
triplets_<video_id>.json       knowledge graph source data (per video)
graph.json                     cumulative graph in node-link format
graph.html                     open in browser — interactive visualization

Each file is named after the video ID, so multiple videos from the same channel are stored without overwriting each other.

Transcription modes

Mode	Speed	Requirement
Fast (default)	Seconds	Video must have subtitles
Whisper (local)	Minutes	Any video, no subtitles needed
WhisperX (opt-in)	Minutes	pip install -r requirements-whisperx.txt

The /process command tries fast mode first and falls back to Whisper automatically. To force Whisper:

python src/transcribe_whisper.py https://www.youtube.com/watch?v=VIDEO_ID

WhisperX mode

WhisperX provides word-level timestamps, speaker diarization, and faster transcription via faster-whisper. Install the extra dependencies:

pip install -r requirements-whisperx.txt

Then use the --engine flag:

/process --engine whisperx https://www.youtube.com/watch?v=VIDEO_ID

Speaker diarization requires a HuggingFace token set as HF_TOKEN in your environment. Without it, WhisperX still produces timestamped segments but without speaker labels.

Obsidian integration

Add --obsidian to export the knowledge graph directly into your Obsidian vault.

/process --obsidian https://www.youtube.com/watch?v=VIDEO_ID

This generates one .md file per extracted entity in vault/content/<channel_name>/obsidian/. Each file contains:

• YAML frontmatter — tags, source title, URL, channel
• Relations — outgoing links to other entities: - made_by: [[Anthropic]]
• Referenced by — incoming links: - [[Claude]] → made_by

Wikilinks use Obsidian's [[filename|display]] aliasing, so they resolve correctly even when entity names contain special characters.

To connect the exported notes to your main vault, point Obsidian at the vault/ folder or copy the obsidian/ directory into your existing vault.

You can also run the exporter directly:

python src/obsidian_exporter.py vault/content/<channel>/triplets_<video_id>.json /path/to/output \
  --metadata vault/content/<channel>/raw/metadata_<video_id>.json

Options

/process [--depth light|standard|deep] [--engine whisperx|whisper] [--obsidian] <URL> [<URL2> ...]

Flag	Values	Default	Effect
--depth	light, standard, deep	standard	Controls triplet count and summary detail
--engine	whisper, whisperx	whisper	Selects transcription engine for fallback
--obsidian	—	off	Exports entity notes to obsidian/ subfolder

Flags apply to all videos in a batch.

Examples:

/process https://www.youtube.com/watch?v=VIDEO_ID
/process https://youtube.com/watch?v=abc https://youtube.com/watch?v=def
/process https://www.youtube.com/playlist?list=PLxxx
/process --depth deep https://www.youtube.com/watch?v=VIDEO_ID
/process --engine whisperx --depth light https://www.youtube.com/watch?v=VIDEO_ID
/process --obsidian https://www.youtube.com/watch?v=VIDEO_ID
/process --obsidian --depth deep https://www.youtube.com/watch?v=VIDEO_ID

License

MIT