# Notebooklm

> Query, manage, and automate Google NotebookLM from your AI agent. Add sources, generate slide decks with your brand colors, run research, and keep auth alive automatically.

- Skill: `robonuggets/notebooklm` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add robonuggets/notebooklm`
- Raw SKILL.md: https://api.skillmd.com/api/skills/robonuggets/notebooklm/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: robonuggets (https://skillmd.com/u/robonuggets)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/robonuggets/notebooklm

---


# NotebookLM Skill by RoboNuggets

Connect your AI agent to Google NotebookLM. Add sources automatically, query your notebooks, generate branded slide decks, and keep your session alive without manual re-login.

Built on top of [notebooklm-py](https://github.com/teng-lin/notebooklm-py) by Teng Lin.

---

## Prerequisites

### 1. Install notebooklm-py

```bash
pip install "notebooklm-py[browser]"
```

The `[browser]` extra installs Playwright, needed for the one-time login. All queries run without a browser after that.

**If pip is unavailable or the package is outdated**, install from a specific GitHub release tag:

```bash
# Get the latest release tag
LATEST_TAG=$(curl -s https://api.github.com/repos/teng-lin/notebooklm-py/releases/latest | grep '"tag_name"' | cut -d'"' -f4)
pip install "git+https://github.com/teng-lin/notebooklm-py@${LATEST_TAG}"
```

> Do NOT install from `git+https://github.com/teng-lin/notebooklm-py` without a tag — the main branch may contain unstable code.

### 2. Install Playwright's Chromium browser (one-time)

```bash
playwright install chromium
```

---

## Auth Setup (One-Time)

```bash
python scripts/nlm.py login
```

This opens a browser window. Sign into your Google account, wait until you see the NotebookLM homepage, then press ENTER in the terminal. Your session is saved to `~/.notebooklm/storage_state.json`.

**After this, your agent can access NotebookLM without any manual login.**

### Verify it worked

```bash
python scripts/nlm.py list
```

You should see your notebooks listed.

---

## Core Commands

All commands run through `python scripts/nlm.py`.

### Notebooks

```bash
# List all notebooks
python scripts/nlm.py list

# Create a new notebook
python scripts/nlm.py create "My Research"

# Describe a notebook (AI summary of contents)
python scripts/nlm.py describe NOTEBOOK_ID
```

### Sources

```bash
# Add a URL (articles, YouTube videos, etc.)
python scripts/nlm.py add-source --notebook-id NOTEBOOK_ID --url "https://example.com/article"

# Add a local file (PDF, markdown, txt, docx)
python scripts/nlm.py add-source --notebook-id NOTEBOOK_ID --file "/path/to/doc.pdf"

# Add raw text
python scripts/nlm.py add-source --notebook-id NOTEBOOK_ID --title "My Notes" --text "Content here..."

# List sources in a notebook
python scripts/nlm.py sources --notebook-id NOTEBOOK_ID
```

### Querying

```bash
# Ask a question (set active notebook first, or pass --notebook-id)
python scripts/nlm.py ask "What are the main concepts?" --notebook-id NOTEBOOK_ID
```

### Local Library (save notebook metadata)

```bash
# Add to local library with a name and description
python scripts/nlm.py library-add --notebook-id NOTEBOOK_ID --name "Claude Docs" --description "All Claude documentation"

# Set as the active (default) notebook
python scripts/nlm.py library-activate claude-docs

# List library
python scripts/nlm.py library-list
```

### Audio & Reports

```bash
# Generate deep-dive audio (podcast style)
python scripts/nlm.py generate-audio --notebook-id NOTEBOOK_ID

# Generate a briefing doc
python scripts/nlm.py generate-report --notebook-id NOTEBOOK_ID

# Generate a study guide
python scripts/nlm.py generate-report --notebook-id NOTEBOOK_ID --format study_guide

# Generate a blog post
python scripts/nlm.py generate-report --notebook-id NOTEBOOK_ID --format blog_post
```

### Artifacts

```bash
# List all generated artifacts (audio, reports, etc.)
python scripts/nlm.py artifacts --notebook-id NOTEBOOK_ID
```

---

## Slide Generation

### Before generating slides — ask the user for their brand colors

**Agent instruction:** Before running any slide generation, always ask the user:

> "What are your brand colors? I need:
> - Primary color (main accent — e.g. `#FF6B2C`)
> - Secondary color (supporting — e.g. `#FFBE2E`)
> - Background color (e.g. `#000000` for black, `#1A1A2E` for dark navy)
> - Text color (e.g. `#F5F5F5` for off-white)
>
> If you don't have hex codes, describe them and I'll pick the closest match."

Once you have the colors, substitute them into the slide prompt template below.

---

### Slide Prompt Template (Blackboard Style)

Use this as a **Focus Prompt** inside NotebookLM (Studio → Focus). The agent fills in the placeholders before sending.

```
Create a [NUMBER]-slide presenter deck for "[VIDEO TITLE]."

Design: Dark blackboard on [BACKGROUND_COLOR]. TITLES: bold slab serif (Rockwell or Roboto Slab), stamped/typeset, color [PRIMARY_COLOR]. Body text: handwritten chalk-style, color [TEXT_COLOR]. Each slide features one key glossy 3D object that floats above the chalk background. Chalk dashed connector lines link 3D elements to surrounding sketch labels. Every slide: dashed border in [PRIMARY_COLOR] + [CORNER_ICON] icon top-right corner. [PRIMARY_COLOR] PRIMARY, [SECONDARY_COLOR] SECONDARY, [TEXT_COLOR] text, [BACKGROUND_COLOR] background.

Slide 1: [SLIDE_1_DESCRIPTION]
Slide 2: [SLIDE_2_DESCRIPTION]
...

CRITICAL: Every slide — slab serif TITLE in [PRIMARY_COLOR] at top, dashed [PRIMARY_COLOR] border, [CORNER_ICON] icon top-right. Chalk body text. [BACKGROUND_COLOR] background. One 3D object per slide. [PRIMARY_COLOR] PRIMARY.
```

**Placeholders to fill:**
| Placeholder | What to fill |
|-------------|-------------|
| `[NUMBER]` | How many slides |
| `[VIDEO TITLE]` | The topic/title of the content |
| `[PRIMARY_COLOR]` | User's primary brand color (hex) |
| `[SECONDARY_COLOR]` | User's secondary brand color (hex) |
| `[BACKGROUND_COLOR]` | Background color (hex) |
| `[TEXT_COLOR]` | Text color (hex) |
| `[CORNER_ICON]` | Icon that appears on every slide (e.g. lightning bolt, star, logo description) |
| `[SLIDE_N_DESCRIPTION]` | Per-slide: what the slide covers, what 3D object appears, what labels surround it |

### Per-slide description format

Each slide description should be ~150 characters. Be specific:

```
Slide 2: [TOPIC]. 3D [OBJECT] floating [POSITION]. Chalk labels: "[LABEL 1]," "[LABEL 2]," "[LABEL 3]." [PRIMARY_COLOR] dashed arrow pointing to [KEY ELEMENT].
```

**3D object suggestions by topic:**
- Data / knowledge → glowing books or filing cabinet
- Actions / automation → terminal / monitor
- Time / scheduling → clock or calendar
- Mobile / notifications → smartphone
- AI / thinking → brain model
- Cloud / remote → glowing cloud
- Steps / process → numbered podium

### How to submit the focus prompt

Once filled in, run:

```bash
python scripts/nlm.py generate-report --notebook-id NOTEBOOK_ID --format custom --prompt "YOUR FILLED PROMPT HERE"
```

Or paste it directly into NotebookLM → Studio → Focus prompt.

> **Note:** NotebookLM has a ~5,000 character limit on focus prompts. If your prompt is too long, shorten the per-slide descriptions.

---

## Keeping Auth Alive (Optional)

Google session cookies expire after 7–30 days. There are two approaches:

### Option A — Simple (re-login when it breaks)

When you get an auth error, just run:

```bash
python scripts/nlm.py login
```

Takes 30 seconds. Fine for occasional use.

### Option B — Headless Auto-Refresh (recommended for agent setups)

`scripts/refresh_auth.py` silently refreshes your cookies using the persistent browser profile — no browser window, no manual steps. Run it every few days before cookies expire.

```bash
python scripts/refresh_auth.py          # Refresh cookies
python scripts/refresh_auth.py --check  # Just check if they're valid
```

**To automate this**, add a scheduled task or cron job:

```
# Cron example — refresh every 3 days at 5:30am
30 5 */3 * * python /path/to/scripts/refresh_auth.py
```

**For Claude Code agent setups**, add this to your agent's `cron-registry.json`:

```json
{
  "id": "notebooklm-auth-refresh",
  "name": "NotebookLM cookie refresh (every 3 days)",
  "cron": "30 5 */3 * *",
  "prompt": "Run: python /path/to/scripts/refresh_auth.py — if it fails, alert the user to run notebooklm login manually.",
  "enabled": true
}
```

> **How it works:** Your first login saves a full browser profile to `~/.notebooklm/browser_profile/`. Google keeps this session alive much longer than raw cookies because the browser handles token rotation automatically. The refresh script uses this profile headlessly to export fresh cookies — no human interaction needed.

---

## Troubleshooting

**Auth error / "Authentication expired"**
Run `python scripts/refresh_auth.py` first. If that fails, run `python scripts/nlm.py login` to do a full re-login.

**`nlm` command not found in terminal**
Use `python scripts/nlm.py` instead. This is a Windows PATH issue with certain Python installations (common with Windows Store Python). The scripts work fine when called with `python` directly.

**"No notebook specified"**
Either pass `--notebook-id YOUR_ID` or set an active notebook first:
```bash
python scripts/nlm.py library-activate your-notebook-name
```

**Slide generation times out**
NotebookLM can take 3–6 minutes to generate custom reports. If the command times out, check your artifacts list — the generation likely completed on NotebookLM's side:
```bash
python scripts/nlm.py artifacts --notebook-id NOTEBOOK_ID
```

**Playwright / Chromium not found**
Run `playwright install chromium` to install the browser.

**Windows: "UnicodeEncodeError" on output**
Set `PYTHONUTF8=1` in your environment, or run `$env:PYTHONUTF8="1"` in PowerShell before running the script.

---

## Credits

- **notebooklm-py** by [Teng Lin](https://github.com/teng-lin/notebooklm-py) — the library that makes this possible
- **This skill** by [RoboNuggets](https://robonuggets.com) — CLI wrapper, headless refresh, slide generation workflow