notebooklm

NotebookLM (Galaxia)

Full programmatic access to Google NotebookLM, including capabilities not exposed in the web UI.

Positioning in the Galaxia stack (read first)

NotebookLM is NOT the agent's memory. The retrieval brain is gbrain + Postgres+pgvector (MCP-native, agent write-back, private). NotebookLM has no official query API: this CLI drives browser automation (cookies), which is fragile and against Google's automated-access terms. So use it for what it is genuinely good at, and never in an autonomous agent's hot path.

Use NotebookLM for:

• Generating media: audio podcast, video explainer, slides, infographic, mind map, quiz, flashcards.
• One-off analysis of a large external corpus (dozens of PDFs/YouTube/URLs) with grounded citations, without building an ingestion pipeline.

Do NOT use NotebookLM for:

• The agent's long-term memory (that is gbrain: gbrain put / the put_page MCP tool).
• Retrieval inside an autonomous agent loop (fragile, ~50 requests/day, breaks on any Google UI change).

Step 0: Setup (on first use)

If notebooklm is not installed or authenticated, do the setup first.

Prerequisite: Python 3.10+

python3 --version

If Python < 3.10 (the macOS default 3.9.x), install a compatible version:

macOS (Homebrew):

brew install [email protected]

Then use /opt/homebrew/bin/python3.12 (Apple Silicon) or /usr/local/bin/python3.12 (Intel).

Linux (apt):

sudo apt update && sudo apt install -y python3.12 python3.12-venv

Install the CLI

Always use a venv (avoids "externally-managed-environment" and PATH issues):

PYTHON=$(command -v python3.12 2>/dev/null || command -v python3.11 2>/dev/null || command -v python3.10 2>/dev/null || command -v python3)
$PYTHON -c "import sys; assert sys.version_info >= (3,10), f'Python {sys.version} too old, need 3.10+'; print(f'Python {sys.version}')"
$PYTHON -m venv ~/.notebooklm-venv
source ~/.notebooklm-venv/bin/activate
pip install "notebooklm-py[browser]"
playwright install chromium

Symlink on PATH:

mkdir -p ~/bin
ln -sf ~/.notebooklm-venv/bin/notebooklm ~/bin/notebooklm
export PATH="$HOME/bin:$PATH"
notebooklm --help

Authenticate

IMPORTANT: notebooklm login needs interactive terminal input (pressing Enter after sign-in), which Claude Code's bash tool does not support. Use this login script instead.

Tell the user:

I am going to open a browser window: sign into your Google account and go to notebooklm.google.com. Take your time, I will wait for your confirmation before closing it.

Then:

cat > /tmp/nlm_login.py << 'PYEOF'
import json, time
from pathlib import Path
from playwright.sync_api import sync_playwright

STORAGE_PATH = Path.home() / ".notebooklm" / "storage_state.json"
PROFILE_PATH = Path.home() / ".notebooklm" / "browser_profile"
SIGNAL_FILE = Path("/tmp/nlm_save_signal")

SIGNAL_FILE.unlink(missing_ok=True)
STORAGE_PATH.parent.mkdir(parents=True, exist_ok=True)

with sync_playwright() as p:
    browser = p.chromium.launch_persistent_context(
        user_data_dir=str(PROFILE_PATH),
        headless=False,
        args=["--disable-blink-features=AutomationControlled"],
    )
    page = browser.pages[0] if browser.pages else browser.new_page()
    page.goto("https://notebooklm.google.com/")
    print("Browser open. Waiting for the save signal...")
    while not SIGNAL_FILE.exists():
        time.sleep(1)
    storage = browser.storage_state()
    with open(STORAGE_PATH, "w") as f:
        json.dump(storage, f)
    print(f"Saved {len(storage.get('cookies', []))} cookies")
    browser.close()

SIGNAL_FILE.unlink(missing_ok=True)
print(f"Auth saved: {STORAGE_PATH}")
PYEOF

source ~/.notebooklm-venv/bin/activate
python3 /tmp/nlm_login.py > /tmp/nlm_login_output.txt 2>&1 &
echo "Login started (PID=$!). Browser opens in a few seconds..."

Wait ~10s, ask the user if they see the browser and are signed in. Once on the NotebookLM home:

touch /tmp/nlm_save_signal
sleep 8
cat /tmp/nlm_login_output.txt

Verify:

export PATH="$HOME/bin:$PATH"
notebooklm auth check
notebooklm list

If OK (SID cookie present), confirm and clean up:

rm -f /tmp/nlm_login.py /tmp/nlm_login_output.txt /tmp/nlm_save_signal

If it fails (SID missing), reset the profile and retry:

rm -rf ~/.notebooklm/browser_profile ~/.notebooklm/storage_state.json

Use from Hermes (Galaxia, optional)

NotebookLM is NOT wired as an MCP server in Hermes (no official query API, browser automation too fragile for a 24/7 agent). To trigger a NotebookLM generation from the autonomous stack, go through N8N (deterministic workflow: receive a request, run the CLI on a machine with a session, deliver the media). Keep NotebookLM out of the agent's hot path.

Autonomy rules

Run without confirmation:

• notebooklm status, auth check, list, source list, artifact list, language list/get/set, artifact wait, source wait, research status/wait, use <id>, create, ask "..." (without --save-as-note), history, source add

Ask before:

• delete (destructive), generate * (long, may fail), download * (writes to disk), ask "..." --save-as-note (writes a note), history --save

Quick reference

Task	Command
List notebooks	notebooklm list
Create notebook	notebooklm create "Title"
Set context	notebooklm use <notebook_id>
Context status	notebooklm status
Add URL	notebooklm source add "https://..."
Add file	notebooklm source add ./file.pdf
Add YouTube	notebooklm source add "https://youtube.com/..."
List sources	notebooklm source list
Wait for source processing	notebooklm source wait <source_id>
Web research (fast)	notebooklm source add-research "query"
Web research (deep)	notebooklm source add-research "query" --mode deep --no-wait
Query	notebooklm ask "question"
Query (specific sources)	notebooklm ask "question" -s src_id1 -s src_id2
Query (with citations)	notebooklm ask "question" --json
Save answer as note	notebooklm ask "question" --save-as-note
Source fulltext	notebooklm source fulltext <source_id>
Generate podcast	notebooklm generate audio "instructions"
Generate video	notebooklm generate video "instructions"
Generate report	notebooklm generate report --format briefing-doc
Generate quiz	notebooklm generate quiz
Generate flashcards	notebooklm generate flashcards
Generate infographic	notebooklm generate infographic
Generate mind map	notebooklm generate mind-map
Generate slides	notebooklm generate slide-deck
Artifact status	notebooklm artifact list
Wait for completion	notebooklm artifact wait <artifact_id>
Download audio	notebooklm download audio ./output.mp3
Download video	notebooklm download video ./output.mp4
Download slides PDF	notebooklm download slide-deck ./slides.pdf
Download report	notebooklm download report ./report.md

Generation types

All generate commands support -s/--source, --language, --json, --retry N.

Type	Command	Options	Output
Podcast	generate audio	--format [deep-dive|brief|critique|debate], --length [short|default|long]	.mp3
Video	generate video	--format [explainer|brief], --style [auto|classic|whiteboard|kawaii|anime|watercolor|retro-print|heritage|paper-craft]	.mp4
Slides	generate slide-deck	--format [detailed|presenter], --length [default|short]	.pdf / .pptx
Infographic	generate infographic	--orientation [landscape|portrait|square], --detail [concise|standard|detailed]	.png
Report	generate report	--format [briefing-doc|study-guide|blog-post|custom], --append "..."	.md
Mind Map	generate mind-map	(instant)	.json
Quiz	generate quiz	--difficulty [easy|medium|hard], --quantity [fewer|standard|more]	.json/.md/.html
Flashcards	generate flashcards	same as quiz	.json/.md/.html

Common workflows

Research to podcast

1. notebooklm create "Research: [topic]"
2. notebooklm source add for each URL/document
3. Wait: notebooklm source list --json until status=ready
4. notebooklm generate audio "Focus on [angle]"
5. notebooklm artifact list for status
6. notebooklm download audio ./podcast.mp3

Large-corpus analysis (where NotebookLM beats gbrain)

1. notebooklm create "Analysis: [project]"
2. notebooklm source add (dozens of PDFs/URLs)
3. notebooklm ask "Synthesize the key points" --json (grounded answers)
4. If durably useful: write the synthesis back into gbrain (gbrain put), do not leave it in NotebookLM.

Error handling

Error	Cause	Action
Auth/cookie	Session expired	Re-login (Step 0 script)
"No notebook context"	Context not set	notebooklm use <id>
Rate limit	Google throttle	Wait 5-10 min
Download fails	Generation incomplete	artifact list

Known limitations

• Audio/video/quiz/flashcards/infographic/slides generation can fail (Google rate limits).
• Timing: audio 10-20 min, video 15-45 min, quiz/flashcards 5-15 min.
• Unofficial API (browser automation): Google can break it without notice. Do not depend on it for a production agent.