name: notebooklm description: Complete API for Google NotebookLM - full programmatic access including features not in the web UI. Create notebooks, add sources, generate all artifact types, download in multiple formats. Activates on explicit /notebooklm or intent like "create a podcast about X"

NotebookLM Automation

Complete programmatic access to Google NotebookLM—including capabilities not exposed in the web UI. Create notebooks, add sources (URLs, YouTube, PDFs, audio, video, images), chat with content, generate all artifact types, and download results in multiple formats.

Installation

From PyPI (Recommended for AI agents — Python-version-aware):

pip install "notebooklm-py[browser]"   # mandatory; errors must propagate

# [cookies] (rookiepy) is optional and known to FAIL TO BUILD on Python 3.13+.
# Skip it deliberately on 3.13+ rather than swallowing the error — that lets
# *real* install failures (typos, network, PyPI outages) surface for the agent.
if python -c "import sys; sys.exit(0 if sys.version_info < (3, 13) else 1)"; then
    pip install "notebooklm-py[cookies]"   # errors propagate
else
    echo "Skipping [cookies] on Python 3.13+ (rookiepy unavailable). Use 'notebooklm login' interactively."
fi

Full install matrix (extras, headless servers, contributor flow): Installation guide on GitHub.

From GitHub (use latest release tag, NOT main branch):

# Get the latest release tag (requires curl + jq)
if ! command -v jq >/dev/null; then
    echo "jq is required to read the latest release tag" >&2
    exit 1
fi
LATEST_TAG=$(
    curl -fsSL https://api.github.com/repos/teng-lin/notebooklm-py/releases/latest |
    jq -r '.tag_name'
)
# Includes [browser] so the interactive `notebooklm login` flow works.
pip install "notebooklm-py[browser] @ git+https://github.com/teng-lin/notebooklm-py@${LATEST_TAG}"

⚠️ DO NOT install from main branch (pip install git+https://github.com/teng-lin/notebooklm-py). The main branch may contain unreleased/unstable changes. Always use PyPI or a specific release tag, unless you are testing unreleased features.

Skill install methods:

• notebooklm skill install installs this skill into the supported local agent directories managed by the CLI.
• npx skills add teng-lin/notebooklm-py installs this skill from the GitHub repository into compatible agent skill directories.
• If you are already reading this file inside an agent skill directory, the skill is already installed. You only need the Python package and authentication below.

CLI-managed install:

notebooklm skill install

Prerequisites

IMPORTANT: Before using any command, you MUST authenticate:

notebooklm login          # Opens browser for Google OAuth
notebooklm list           # Verify authentication works

If commands fail with authentication errors, re-run notebooklm login.

CI/CD, Multiple Accounts, and Parallel Agents

For automated environments, multiple accounts, or parallel agent workflows:

CI/CD setup: Set NOTEBOOKLM_AUTH_JSON from a secret containing your storage_state.json contents.

Multiple accounts: Use named profiles (notebooklm profile create work, then notebooklm -p work login). Alternatively, use different NOTEBOOKLM_HOME directories per account.

Parallel agents: The CLI stores notebook context per profile (~/.notebooklm/profiles/<profile>/context.json, with a legacy fallback to ~/.notebooklm/context.json for the implicit default profile). Multiple concurrent agents that share a profile and use notebooklm use can overwrite each other's context — use one of the isolation strategies below.

Solutions for parallel workflows:

1. Always use explicit notebook ID (recommended): Pass -n <notebook_id> / --notebook <notebook_id> on notebook-scoped commands instead of relying on use
2. Per-agent isolation via profiles: export NOTEBOOKLM_PROFILE=agent-$ID (each profile gets its own context file)
3. Per-agent isolation via home: Set unique NOTEBOOKLM_HOME per agent: export NOTEBOOKLM_HOME=/tmp/agent-$ID
4. Use full UUIDs: Avoid partial IDs in automation (they can become ambiguous)

Agent Setup Verification

Before starting workflows, verify auth is in place. Use --test --json (not bare --json) — bare --json only proves the cookie file parses; --test makes a network call and proves the cookies still authenticate against Google.

1. notebooklm auth check --test --json → require BOTH "status": "ok" AND "checks.token_fetch": true. Bare "status": "ok" (without --test) is a false-positive trap — a stale cookie file passes the parse check.
2. notebooklm list --json → expect valid JSON (may be empty for new accounts).
3. If auth fails or is missing → run notebooklm login first. This is the primary auth path: opens a browser, the user signs in to Google once, and the resulting storage_state.json is reused on every subsequent run. Works on any environment with a display.
   • For headless contexts where opening a browser is not feasible, use notebooklm login --browser-cookies <browser> instead — extracts the user's already-logged-in cookies from Chrome/Firefox/etc. (requires the [cookies] extra; rookiepy may not install on Python 3.13+). Use chrome::<profile-name-or-directory> to target one Chromium user-profile, or firefox::<container-name> / firefox::none to target one Firefox container.
   • To survey signed-in Google accounts before picking one: notebooklm auth inspect --browser <browser> (read-only; pass -v to see which Chromium user-profile each account came from, or --json for tooling). Scoped forms such as notebooklm auth inspect --browser 'chrome::Profile 1' inspect only that browser profile.
   • Re-run step 1 after login to confirm.
4. If auth was working but cookies went stale (Google rotated SIDTS, or you signed in fresh in the browser) → refresh the active profile in place instead of full re-login:
   • notebooklm auth refresh — server-side SIDTS refresh against the existing storage_state.json. Cheap and silent; safe to run on a schedule (cron / launchd / systemd) at 15–20 min cadence to keep an unattended profile warm.
   • notebooklm auth refresh --browser-cookies <browser> — re-extract cookies from a running browser and match them back to the profile's recorded email in context.json. Use when the on-disk storage_state.json is too stale for the server-side refresh path but you've just signed back into Google in the browser. For Chromium-family browsers with multiple user-profiles (Chrome's Default, Profile 1, …), refresh fans out across all profiles to find the email — same path as auth inspect (issue #571). Use chrome::<profile-name-or-directory> when you already know the exact browser profile.
   • Both forms preserve the same --profile (no new profile is created).

Note: notebooklm status reports context state (selected notebook); do not use it to verify auth.

When This Skill Activates

Explicit: User says "/notebooklm", "use notebooklm", or mentions the tool by name

Intent detection: Recognize requests like:

• "Create a podcast about [topic]"
• "Summarize these URLs/documents"
• "Generate a quiz from my research"
• "Turn this into an audio overview"
• "Create flashcards for studying"
• "Generate a video explainer"
• "Make an infographic"
• "Create a mind map of the concepts"
• "Download the quiz as markdown"
• "Add these sources to NotebookLM"

Autonomy Rules

Run automatically (no confirmation):

• notebooklm status - check context
• notebooklm auth check - diagnose auth issues
• notebooklm auth inspect - list Google accounts visible to a browser (read-only)
• notebooklm auth refresh - server-side SIDTS refresh of the active profile (no new profile, no destructive writes)
• notebooklm auth refresh --browser-cookies <browser> - re-extract cookies from a browser into the active profile (rebuilds storage_state.json for the same --profile, not a new one)
• notebooklm list - list notebooks
• notebooklm source list - list sources
• notebooklm artifact list - list artifacts
• notebooklm language list - list supported languages
• notebooklm language get - get current language
• notebooklm language set - set language (global setting)
• notebooklm artifact wait - wait for artifact completion (in subagent context)
• notebooklm source wait - wait for source processing (in subagent context)
• notebooklm research status - check research status
• notebooklm research wait - wait for research (in subagent context)
• notebooklm use <id> - set context (⚠️ SINGLE-AGENT ONLY - use -n flag in parallel workflows)
• notebooklm create - create notebook
• notebooklm ask "..." - chat queries (without --save-as-note)
• notebooklm suggest-prompts - AI-suggested prompts for a notebook (read-only, no state change)
• notebooklm history - display conversation history (read-only)
• notebooklm source add - add sources
• notebooklm profile list - list profiles
• notebooklm profile create - create profile
• notebooklm profile switch - switch active profile
• notebooklm doctor - check environment health

Ask before running:

• notebooklm delete, source delete, source delete-by-title, source clean, note delete, artifact delete, label delete, share remove, auth logout, clear, profile delete, or ask --new - destructive or state-changing. Once approved, pass --yes/-y where the command supports it. Most destructive --json commands still require explicit --yes and otherwise return a structured confirmation error (CONFIRM_REQUIRED or VALIDATION_ERROR, depending on the command family); current exceptions include share remove --json and ask --new --json, which skip the prompt for non-interactive callers.
• notebooklm generate * - long-running, may fail
• notebooklm download * - writes to filesystem
• notebooklm artifact wait - long-running (when in main conversation)
• notebooklm source wait - long-running (when in main conversation)
• notebooklm research wait - long-running (when in main conversation)
• notebooklm research cancel <run_id> - state-changing; cancels a running research job (an in-progress job transitions to FAILED). Fire-and-forget: it does not confirm success — re-check with notebooklm research status.
• notebooklm ask "..." --save-as-note - writes a note
• notebooklm history --save - writes a note

Quick Reference

Parallel safety: Use explicit notebook IDs in parallel workflows. Notebook-scoped commands broadly support -n/--notebook (ask/history, source, artifact, generate, download, note, label, share, research, and notebook delete/rename/summary/metadata). Download commands also support -a/--artifact. For chat, use -c <conversation_id> to target a specific conversation.

Partial IDs: Use first 6+ characters of UUIDs. Must be unique prefix (fails if ambiguous). Works for ID-based commands such as use, source delete, and wait. For exact source-title deletion, use source delete-by-title "Title". For automation, prefer full UUIDs to avoid ambiguity.

Command Output Formats

Commands with --json return structured data for parsing:

Create notebook:

$ notebooklm create "Research" --json
{"notebook": {"id": "abc123de-...", "title": "Research", "created_at": null}}
# parse with: jq -r .notebook.id

Add source:

$ notebooklm source add "https://example.com" --json
{"source": {"id": "def456...", "title": "Example", "type": "web_page", "url": "https://example.com"}}
# parse with: jq -r .source.id
# Note: no `status` field on add — use `source list --json` or `source wait` to check processing state.

Generate artifact:

$ notebooklm generate audio "Focus on key points" --json
{"task_id": "xyz789...", "status": "pending"}
# When run with --wait, completed status also includes a `url` field.

Chat with references:

$ notebooklm ask "What is X?" --json
{"answer": "X is... [1] [2]", "conversation_id": "...", "turn_number": 1, "is_follow_up": false, "references": [{"source_id": "abc123...", "citation_number": 1, "cited_text": "Relevant passage from source..."}, {"source_id": "def456...", "citation_number": 2, "cited_text": "Another passage..."}]}

Source fulltext (get indexed content):

$ notebooklm source fulltext <source_id> --json
{"source_id": "...", "title": "...", "kind": "web_page", "content": "Full indexed text...", "url": null, "char_count": 12345}

Understanding citations: The cited_text in references is often a snippet or section header, not the full quoted passage. The start_char/end_char positions reference NotebookLM's internal chunked index, not the raw fulltext. Use SourceFulltext.find_citation_context() to locate citations:

fulltext = await client.sources.get_fulltext(notebook_id, ref.source_id)
matches = fulltext.find_citation_context(ref.cited_text)  # Returns list[(context, position)]
if matches:
    context, pos = matches[0]  # First match; check len(matches) > 1 for duplicates

Extract IDs: Singular endpoints wrap their result in an envelope — parse .notebook.id (from create), .source.id (from source add), or .task_id (from generate *). The chat --json references list uses .references[].source_id.

Generation Types

Common generate options vary by subcommand:

• -n, --notebook targets the notebook.
• -s, --source limits generation to specific source(s) on content generators (not revise-slide).
• --language sets output language where supported (defaults to configured language or en).
• --wait, --timeout, and --interval are shared polling controls where waiting is supported.
• --json returns machine-readable output.
• --retry N automatically retries rate limits on supported subcommands (not mind-map).
• --prompt-file PATH reads description/query text from a file on ask, generation subcommands except mind-map, and source add-research.

¹ --append only customizes the built-in templates. With --format custom, pass the prompt as the positional DESCRIPTION argument (notebooklm generate report "PROMPT" --format custom); --append is silently ignored in that mode (the CLI prints a warning).

³ Two kinds of mind map (issue #1256). generate mind-map --kind interactive (the default) creates the interactive studio artifact (what the web app now makes); it is polled to completion. generate mind-map --kind note-backed creates the note-backed kind — a JSON node tree, generated synchronously. Both emit the same {mind_map, note_id, kind} JSON, list under artifact list --type mind-map, and export via download mind-map. --instructions applies only to the note-backed kind.

⁴ Cinematic video (Veo 3). generate video --format cinematic generates AI documentary footage via Veo 3; it ignores --style, takes ~30-40 min, and requires a Google AI Ultra subscription. Also exposed as the generate cinematic-video alias (which forces --format cinematic and a longer default timeout). Download with download video or the download cinematic-video alias.

² Portrait / vertical slide decks via prompt. Slide-deck has no --orientation flag (unlike infographic). Treat portrait decks as skill-level prompt guidance, not a typed CLI/API contract: NotebookLM currently honors orientation cues written into the DESCRIPTION positional argument. Including phrases like "9:16 portrait", "vertical layout", "portrait mobile format", or "vertical 9:16 layout" can make NotebookLM render each slide as a 9:16 portrait image. Empirically:

• The .pptx canvas itself may stay 16:9, but each slide's embedded image can be rendered as 9:16 portrait — useful for vertical/mobile video material extracted via python-pptx.
• Orientation is steered once at generation time. generate revise-slide edits content within an existing slide but does not change its orientation; if a slide falls back to landscape (occasional inconsistency), regenerate the whole deck rather than revising the single page.
• Combine with an explicit page count in the prompt (e.g. "Create exactly 8 pages, using a vertical 9:16 portrait layout") for the most predictable output.

# Skill prompt hint: ask NotebookLM to render each slide as a 9:16 portrait image
notebooklm generate slide-deck "Create an 8-page deck in 9:16 portrait orientation for mobile viewing" --length default

Features Beyond the Web UI

These capabilities are available via CLI but not in NotebookLM's web interface:

Common Workflows

Research to Podcast (Interactive)

Time: 5-10 minutes total

1. notebooklm create "Research: [topic]" — if fails: check auth with notebooklm login
2. notebooklm source add for each URL/document — if one fails: log warning, continue with others
3. Wait for sources: notebooklm source list --json until all status=READY — required before generation
4. notebooklm generate audio "Focus on [specific angle]" (confirm when asked) — if rate limited: wait 5 min, retry once
5. Note the artifact ID returned
6. Check notebooklm artifact list later for status
7. notebooklm download audio ./podcast.mp3 when complete (confirm when asked)

Research to Podcast (Automated with Subagent)

Time: 5-10 minutes, but continues in background

When user wants full automation (generate and download when ready):

1. Create notebook and add sources as usual
2. Wait for sources to be ready (use source wait or check source list --json)
3. Run notebooklm generate audio "..." --json → parse task_id from output
4. Spawn a background agent using Task tool:

   Task(
     prompt="Wait for artifact {task_id} in notebook {notebook_id} to complete, then download.
             Use: notebooklm artifact wait {task_id} -n {notebook_id} --timeout 1200
             Then: notebooklm download audio ./podcast.mp3 -a {task_id} -n {notebook_id}",
     subagent_type="general-purpose"
   )
   

5. Main conversation continues while agent waits

Error handling in subagent:

• If artifact wait returns exit code 2 (timeout): Report timeout, suggest checking artifact list
• If download fails: Check if artifact status is COMPLETED first

Benefits: Non-blocking, user can do other work, automatic download on completion

Document Analysis

Time: 1-2 minutes

1. notebooklm create "Analysis: [project]"
2. notebooklm source add ./doc.pdf (or URLs)
3. notebooklm ask "Summarize the key points"
4. notebooklm ask "What are the main arguments?"
5. Continue chatting as needed

Bulk Import

Time: Varies by source count

1. notebooklm create "Collection: [name]"
2. Add multiple sources:

   notebooklm source add "https://url1.com"
   notebooklm source add "https://url2.com"
   notebooklm source add ./local-file.pdf
   

3. notebooklm source list to verify

Source limits: Varies by plan—Standard: 50, Plus: 100, Pro: 300, Ultra: 600 sources per notebook. See NotebookLM plans for details. The CLI does not enforce these limits; they are applied by your NotebookLM account. Supported types: PDFs, YouTube URLs, web URLs, Google Docs, text files, Markdown, Word docs, EPUB, audio files, video files, images

Bulk Import with Source Waiting (Subagent Pattern)

Time: Varies by source count

When adding multiple sources and needing to wait for processing before chat/generation:

1. Add sources with --json to capture IDs (parse with jq -r .source.id):

   notebooklm source add "https://url1.com" --json  # → {"source": {"id": "abc...", ...}}
   notebooklm source add "https://url2.com" --json  # → {"source": {"id": "def...", ...}}
   

2. Spawn a background agent to wait for all sources:

   Task(
     prompt="Wait for sources {source_ids} in notebook {notebook_id} to be ready.
             For each: notebooklm source wait {id} -n {notebook_id} --timeout 600
             Report when all ready or if any fail.",
     subagent_type="general-purpose"
   )
   

3. Main conversation continues while agent waits
4. Once sources are ready, proceed with chat or generation

Why wait for sources? Sources must be indexed before chat or generation. Takes ~30 seconds to several minutes per source (see the processing-times table below).

Deep Web Research (Subagent Pattern)

Time: 15-30+ minutes, runs in background

Deep research finds and analyzes web sources on a topic:

1. Create notebook: notebooklm create "Research: [topic]"
2. Start deep research (non-blocking):

   notebooklm source add-research "topic query" --mode deep --no-wait
   

3. Spawn a background agent to wait and import:

   Task(
     prompt="Wait for research in notebook {notebook_id} to complete and import sources.
             Use: notebooklm research wait -n {notebook_id} --import-all --timeout 1800
             Report how many sources were imported.",
     subagent_type="general-purpose"
   )
   

4. Main conversation continues while agent waits
5. When agent completes, sources are imported automatically

Alternative (blocking): For simple cases, omit --no-wait:

notebooklm source add-research "topic" --mode deep --import-all
# Blocks until research completes (deep mode: 15-30+ min)

When to use each mode:

• --mode fast: Specific topic, quick overview needed (5-10 sources, seconds)
• --mode deep: Broad topic, comprehensive analysis needed (20+ sources, 15-30+ min)

Research sources:

• --from web: Search the web (default)
• --from drive: Search Google Drive

Output Style

Progress updates: Brief status for each step

• "Creating notebook 'Research: AI'..."
• "Adding source: https://example.com..."
• "Starting audio generation... (task ID: abc123)"

Fire-and-forget for long operations:

• Start generation, return artifact ID immediately
• Do NOT poll or wait in main conversation - generation takes 5-45 minutes (see timing table)
• User checks status manually, OR use subagent with artifact wait

JSON output: Use --json flag for machine-readable output:

notebooklm list --json
notebooklm auth check --test --json   # use --test for network-validated auth (see § Agent Setup Verification)
notebooklm source list --json
notebooklm artifact list --json

JSON schemas (key fields):

notebooklm list --json:

{"notebooks": [{"index": 1, "id": "...", "title": "...", "is_owner": true, "created_at": "..."}], "count": 1}

notebooklm auth check --test --json (use --test to drive the network token-fetch — bare --json would leave "token_fetch": null):

{"status": "ok", "checks": {"storage_exists": true, "json_valid": true, "cookies_present": true, "sid_cookie": true, "token_fetch": true}, "details": {"storage_path": "...", "auth_source": "file", "cookies_found": ["SID", "HSID", "..."], "cookie_domains": [".google.com"]}}

notebooklm source list --json:

{"notebook_id": "...", "notebook_title": "...", "sources": [{"index": 1, "id": "...", "title": "...", "type": "web_page", "url": "...", "status": "ready|processing|error", "status_id": 1, "created_at": "..."}], "count": 1}

notebooklm artifact list --json:

{"notebook_id": "...", "notebook_title": "...", "artifacts": [{"index": 1, "id": "...", "title": "...", "type": "Audio", "type_id": 1, "status": "in_progress|pending|completed|unknown", "status_id": 1, "created_at": "..."}], "count": 1}

Status values:

• Sources: processing → ready (or error)
• Artifacts: pending or in_progress → completed (or unknown)

Error Handling

On failure, offer the user a choice:

1. Retry the operation
2. Skip and continue with something else
3. Investigate the error

Error decision tree:

Exit Codes

All commands use consistent exit codes:

Examples:

• source wait returns 1 if source not found or processing failed
• artifact wait returns 2 if timeout reached before completion
• generate returns 1 if rate limited (check stderr for details)

Long Prompts

When a prompt or query exceeds shell command-line length limits, use --prompt-file to read it from a file:

notebooklm ask --prompt-file ./long_question.txt
notebooklm generate report --prompt-file ./custom_report_prompt.txt
notebooklm source add-research --prompt-file ./research_query.txt --mode deep

--prompt-file is mutually exclusive with the positional text argument. The file is read as UTF-8 with trailing whitespace stripped. Supported on: ask, all generate subcommands (except mind-map), and source add-research.

Note: --prompt-file reads a prompt/query text file, not a source document. To upload a file as a notebook source, use source add ./file.pdf.

Known Limitations

Rate limiting: Audio, video, quiz, flashcards, infographic, and slide deck generation may fail due to Google's rate limits. This is an API limitation, not a bug.

Reliable operations: These always work:

• Notebooks (list, create, delete, rename)
• Sources (add, list, delete)
• Chat/queries
• Mind-map, study-guide, report, data-table generation

Unreliable operations: These may fail with rate limiting:

• Audio (podcast) generation
• Video generation
• Quiz and flashcard generation
• Infographic and slide deck generation

Workaround: If generation fails:

1. Check status: notebooklm artifact list
2. Retry after 5-10 minutes
3. Use the NotebookLM web UI as fallback

Processing times vary significantly. Use the subagent pattern for long operations:

Polling intervals: When checking status manually, poll every 15-30 seconds to avoid excessive API calls.

Language Configuration

Language setting controls the output language for generated artifacts (audio, video, etc.).

Important: Language is a GLOBAL setting that affects all notebooks in your account.

# List all 80+ supported languages with native names
notebooklm language list

# Show current language setting
notebooklm language get

# Set language for artifact generation
notebooklm language set zh_Hans  # Simplified Chinese
notebooklm language set ja       # Japanese
notebooklm language set en       # English (default)

Common language codes:

Override per command: Use --language flag on generate commands:

notebooklm generate audio --language ja   # Japanese podcast
notebooklm generate video --language zh_Hans  # Chinese video

Offline mode: Use --local flag to skip server sync:

notebooklm language set zh_Hans --local  # Save locally only
notebooklm language get --local  # Read local config only

Troubleshooting

notebooklm --help              # Main commands
notebooklm auth check          # Diagnose auth issues
notebooklm auth check --test   # Full auth validation with network test
notebooklm source --help       # Source management
notebooklm research --help     # Research status/wait/cancel
notebooklm generate --help     # Content generation
notebooklm artifact --help     # Artifact management
notebooklm download --help     # Download content
notebooklm language --help     # Language settings

Diagnose auth: notebooklm auth check - shows cookie domains, storage path, validation status Re-authenticate: notebooklm login Check version: notebooklm --version Refresh a CLI-managed install: notebooklm skill install

notebooklm-py

A Comprehensive NotebookLM Skill & Unofficial Python API. Full programmatic access to NotebookLM's features—including capabilities the web UI doesn't expose—via Python, CLI, and AI agents like Claude Code, Codex, and OpenClaw.

Source & Development: https://github.com/teng-lin/notebooklm-py

⚠️ Unofficial Library - Use at Your Own Risk

This library uses undocumented Google APIs that can change without notice.

• Not affiliated with Google - This is a community project
• APIs may break - Google can change internal endpoints anytime
• Rate limits apply - Heavy usage may be throttled

Best for prototypes, research, and personal projects. See Troubleshooting for debugging tips.

What You Can Build

🤖 AI Agent Tools - Integrate NotebookLM into Claude Code, Codex, and other LLM agents. Ships with a root NotebookLM skill for GitHub and npx skills add discovery, local notebooklm skill install support for Claude Code and .agents skill directories, and repo-level Codex guidance in AGENTS.md.

📚 Research Automation - Bulk-import sources (URLs, PDFs, YouTube, Google Drive), run web/Drive research queries with auto-import, and extract insights programmatically. Build repeatable research pipelines.

🎙️ Content Generation - Generate Audio Overviews (podcasts), videos, slide decks, quizzes, flashcards, infographics, data tables, mind maps, and study guides. Full control over formats, styles, and output.

📥 Downloads & Export - Download all generated artifacts locally (MP3, MP4, PDF, PNG, CSV, JSON, Markdown). Export to Google Docs/Sheets. Features the web UI doesn't offer: batch downloads, quiz/flashcard export in multiple formats, mind map JSON extraction.

Use Cases & Recipes

NotebookLM is a grounded engine: Gemini does the heavy reading and answers from your sources with citations. The winning pattern is to let it do the expensive analysis while your agent (Claude Code, Codex, …) orchestrates and handles the final mile — using NotebookLM as a zero-token synthesis + memory layer an agent drives in a loop, and pulling structured artifacts out in bulk and in richer, scriptable formats. Recipes people build on top of this library, grouped by what they use NotebookLM as:

Spend fewer tokens — let NotebookLM do the expensive thinking:

• 🪙 Zero-token research offload — Throw 30 documents into a notebook, let Gemini do the heavy analysis, and have your agent spend tokens only on the final polish. The agent just orchestrates (create → source add → ask); the reasoning happens server-side. In the wild: a four-workflow guide to stop Claude Code burning tokens on NotebookLM.
• 🧠 Knowledge distillation → a permanent skill — Run Deep Research (source add-research "your topic" --mode deep) or load a doc corpus, let NotebookLM's Gemini condense it, and bake the result into a SKILL.md your agent loads at startup — build once, reuse with zero runtime tokens or network calls, git-versioned and immune to UI drift. A packaged domain expert without hand-curating sources. (Dumping raw docs into a skill flattens the hierarchy; NotebookLM condensing first is what makes it work.)
• ✅ Self-validating skills — Have NotebookLM generate the eval set — a quiz straight from your sources — to grade an agent skill against ground truth instead of test questions you'd bias yourself. Build the skill, run it against the NotebookLM-authored evals, iterate to a pass. In the wild: a skill that scored 4/10 on the first pass and 10/10 after one iteration, graded by a NotebookLM-generated quiz.

Give your agent memory — persistent, grounded recall:

• 💾 Persistent cross-session memory — Keep a "Master Brain" notebook; a wrap-up step appends each session's decisions and fixes as notes (note create / ask --save-as-note), and a line in your CLAUDE.md queries it (ask) at the start of the next session. Storage and recall live on Google's infrastructure.
• 🧩 Grounded memory for coding agents — Expose a notebook of your internal docs/RFCs/architecture over the MCP server (or plain ask) so an agent answers from your code with citations rather than plausible-sounding guesses — a zero-infra alternative to standing up your own vector DB and embedding pipeline. In the wild: turning a notebook into the source-grounded "project brain" a coding agent consults before it writes code.
• 🪞 Query your own notes / journal — Load years of daily notes, meeting logs, or a journal and ask for cited answers across your own history — surfacing long-term patterns a keyword search can't (e.g. a weekly summary synthesized from 282 daily notes, every claim linked back to the entry it came from). In the wild: chatting with a year of daily notes as a cited knowledge base.

Turn your sources into answers & artifacts — cited responses, generated media, and exports:

• 📞 Grounded knowledge base / troubleshooting oracle (RAG) — Load product docs, FAQs, RFCs, and past tickets, then ask --json for source-grounded, cited answers for support, on-call, or internal Q&A. Or have an agent point it at an entire fast-moving tool's docs — more than the agent can hold in context — as a troubleshooting oracle it queries the moment it hits an error. In the wild: OpenClaw drove the library to scrape all 524 pages of docs.openclaw.ai, dedupe the duplicate translations, and audit it down to 269 clean sources (missing/extra/duplicate = 0).
• 🔁 Multi-format content repurposing — One source set, every format: generate audio (podcast), generate video, generate slide-deck, plus a generate report blog draft, generate quiz, and generate flashcards — fan a single notebook out across channels.
• 📤 Bulk, scriptable exports — Pull mind maps as JSON, flashcards/quizzes as JSON/Markdown/HTML, data tables as CSV, and reports as Markdown — in bulk, to local files, straight into Anki, your mind-mapping tool, or a repo (download <type> / download <type> --all). The programmatic "get data out" half of the library, not just "put sources in."
• 🕸️ Obsidian / knowledge-graph sync — Run the CLI from your vault root so downloaded artifacts (reports, mind-map JSON, transcripts) land as files in your knowledge graph; community skills built on this library even resolve NotebookLM's citation markers into Obsidian [[wikilinks]]. Pair with a podcast overview for an audio digest of your notes. In the wild: "Claude Code + NotebookLM + Obsidian = GOD MODE".

Run it unattended, at scale, or on the go — scheduled, headless, and remote:

• 🚨 Incident runbook generator — On an alert, spin up a notebook of the relevant docs, ask targeted diagnostic questions, and generate a briefing-doc report (generate report --format briefing-doc --wait, then download report) as an automated runbook.
• 📚 Curriculum / study-set builder — Scrape a syllabus or developer roadmap, create one notebook per topic (with deliberate pacing to dodge rate limits), and bulk-generate podcasts, quizzes, and flashcards for each.
• 📰 Scheduled audio briefings — Pair auth refresh --quiet (cron/launchd/systemd) with generate audio to publish a fresh personalized briefing to a podcast feed on a schedule.
• 📱 NotebookLM from your phone, agent-driven — Self-host the remote MCP connector behind a Cloudflare/Tailscale tunnel and add it as a custom connector on the web (claude.ai Connectors, or ChatGPT with Developer Mode). Then drive the full toolset — deep research, source ingestion, studio generation, cited Q&A — from the claude.ai mobile app on the go (ChatGPT's MCP connectors are web-only), chained with your other MCP tools, instead of app-hopping.

These combine ordinary library primitives — see the CLI Reference and Python API. The agent-side glue (skills, scheduling, vault layout) lives in your own setup, not this package. Per-notebook source counts depend on your Google account tier — split across notebooks if you hit a cap.

New here? Start with a walkthrough: Claude Code + NotebookLM = CHEAT CODE (video) · 5 demos + 50 use cases, with prompts.

Ways to Use