Notebooklm

NotebookLM - Grounded Research Projection

Use NotebookLM as a research engine, then project the results into durable markdown.

Current reference projection target is an Obsidian vault. Architecturally, that vault is a publisher/projection adapter, not an authority plane. The valuable part of this skill is the workflow:

1. curate sources
2. load them into NotebookLM
3. ask grounded questions
4. extract cited passages
5. resolve citations into durable local links
6. publish/project the result

Prerequisites

1. Install notebooklm-py

pip install "notebooklm-py[browser]"
playwright install chromium

2. Authenticate

notebooklm login

Cookies are stored at ~/.notebooklm/storage_state.json.

3. Projection Target

The reference target is a markdown vault with Dataview support.

• Obsidian + Dataview is the default adapter
• Other CMS/publisher targets can consume the same markdown artifacts later

4. Optional Templates

Reference templates live in templates/:

• templates/notebook-source.md
• templates/dashboard.md

Quick Start

notebooklm status
notebooklm list
notebooklm use <notebook-id>

Workflow Routing

Reference Projection Layout

Your Vault
├── Notes/NotebookLM/{notebook-slug}/
│   ├── Sources/
│   │   ├── Video Title.md
│   │   └── Article Title.md
│   └── QA/
│       └── 2026-03-07 Emerging Themes.md
└── Notes/Dashboards/
    └── Research Dashboard.md

• Source files carry notebook/source metadata plus extracted guides/topics
• Q&A notes resolve [N] markers to passage-level wikilinks
• Dashboards/materialized views sit on top of the markdown output

Data Model

notebook-source frontmatter:

type: notebook-source
source_id: "uuid"
notebook_id: "uuid"
url: ""
source_type: youtube
status: active
date: YYYY-MM-DD
topics:
  - "[[Topic Name]]"
related:
  - "[[Notes/Dashboards/Dashboard Name]]"

Q&A reference frontmatter:

type: reference
status: current
date: YYYY-MM-DD
source: "notebooklm:{notebook-slug}"
related:
  - "[[Notes/Dashboards/Dashboard Name]]"

Scripts

All scripts use Path.cwd() as the projection root. Run them from the target vault/workspace.