Turn Any YouTube Channel into a NotebookLM Knowledge Base

NotebookLM is brilliant for chatting with documents, but it has one stupid limitation: you can't feed it a YouTube channel. You paste one video at a time, with a 50-source cap, and most channels have hundreds.

So I built youtube-to-notebooklm — a tiny CLI that turns any channel into a folder of clean Markdown files that drag straight into NotebookLM.

@channel  →  yt-dlp lists every upload
          →  transcripts pulled (no API key)
          →  packed into NotebookLM-ready Markdown
          →  drag-and-drop into your notebook

Two output flavours:

• bundles/*.md — many videos packed per file (≤450k words) so a 200-video channel becomes ~3-4 NotebookLM sources, not 200.
• videos/*.md — one file per video. Slower to upload, but every NotebookLM citation links back to a single video.

Install

git clone https://github.com/melyx-id/youtube-to-notebooklm.git
cd youtube-to-notebooklm
pip install -r requirements.txt

Or install as a CLI:

pip install -e .

Requires Python 3.9+. No API key.

Use it in 3 steps

1. Run the tool

# By channel handle
python main.py https://www.youtube.com/@veritasium

# By @handle
python main.py "@lexfridman"

# By playlist
python main.py "https://www.youtube.com/playlist?list=PLrAXtmRdnEQy..."

# Quick test — first 10 videos only
python main.py @veritasium --limit 10

Output lands in notebooklm_export/:

notebooklm_export/
├── INDEX.md          # which videos had transcripts
├── data.json         # structured data for your own pipelines
├── videos/*.md       # one file per video
└── bundles/*.md      # packed bundles for NotebookLM

2. Drag into NotebookLM

1. Open NotebookLM and create a new notebook.
2. Click Add source → upload bundles/*.md (fewer files, full coverage).
3. Wait ~30 seconds for indexing.

3. Ask, summarise, generate

• Ask in any language — NotebookLM translates on the fly.
• From the Studio panel, generate Audio Overviews (two-host podcast), Mind Maps, or Study Guides.
• Need precise per-video citations? Upload videos/*.md instead of bundles.

Useful flags

Programmatic use (Python)

from yt2nlm.fetcher import build_video, list_channel_videos
from yt2nlm.exporter import write_bundles
from pathlib import Path

entries = list_channel_videos("@veritasium", limit=20)
videos  = [build_video(e, languages=["en", "vi"]) for e in entries]
write_bundles(videos, Path("out/bundles"))

Useful when you want to plug it into your own RAG / content pipeline rather than upload to NotebookLM.

Honest caveats

• Caption-only. Videos without auto or manual captions are skipped — see INDEX.md for the missing list.
• Big channels (>500 videos) — use --limit and chunk by year to avoid YouTube rate-limiting.
• Member-only or private videos aren't supported, by design.
• No translation. The tool grabs the original transcript. Let NotebookLM or your favourite LLM translate downstream.

Beyond NotebookLM

NotebookLM is great for Q&A. Once data.json exists, a much bigger world opens up:

• Bulk-summarise an entire 200-video channel in one pass.
• Auto-generate blog posts, X threads, or YouTube Shorts from old videos.
• Build your own RAG with pgvector or Qdrant.
• Run an agent that watches a channel and posts a weekly digest.

I run that pipeline through AIGateCloud — a multi-provider gateway with automatic fallback across OpenAI, Anthropic, Gemini, DeepSeek, and Groq. One API, never blocked by a single provider going down.

Cheatsheet

# Setup once
git clone https://github.com/melyx-id/youtube-to-notebooklm.git
cd youtube-to-notebooklm
pip install -r requirements.txt

# Run
python main.py @CHANNEL_HANDLE --limit 50

# Drag bundles/*.md into NotebookLM. Done.

Repo: github.com/melyx-id/youtube-to-notebooklm — MIT licensed. Stars and PRs welcome.