notebooklm-api

Unofficial TypeScript/JavaScript client and CLI for the Google NotebookLM API.

Fork of notebooklm with extensive bug fixes (15+ RPC fixes) and new features (slide deck generation, artifact downloads). See CHANGELOG.md for details.

Install

npm install -g notebooklm-api

Authentication

notebooklm login

Opens a browser to authenticate with your Google account. Credentials are stored locally in ~/.notebooklm/.

CLI Usage

Notebooks

notebooklm create "My Research"           # Create notebook
notebooklm list                           # List all notebooks

Sources

notebooklm source add-text <notebookId> -t "Title" -c "Content"
notebooklm source add-url <notebookId> <url>
notebooklm source list <notebookId>
notebooklm source delete <notebookId> <sourceId>

Generate Content

# Reports
notebooklm generate report <notebookId> -p "Custom prompt" -w

# Slide decks
notebooklm generate slide-deck <notebookId> -p "Instructions" -f presenter -w

# Audio/video overviews
notebooklm generate audio <notebookId> -w
notebooklm generate video <notebookId> -w

# Check status
notebooklm generate status <notebookId> <artifactId>

# Download content
notebooklm generate download <notebookId> <artifactId> output.md

Research

notebooklm research fast <notebookId> "research query"
notebooklm research deep <notebookId> "research query"

Programmatic Usage

import { NotebookLMClient } from 'notebooklm-api';

const client = await NotebookLMClient.fromStorage();

// Create notebook and add sources
const notebook = await client.notebooks.create('Research');
await client.sources.addText(notebook.id, 'Title', 'Content...');

// Generate report
const status = await client.artifacts.generateReport(notebook.id, {
  customPrompt: 'Analyze the key findings...'
});

// Wait and download
await client.artifacts.waitForGeneration(notebook.id, status.artifactId);
const content = await client.artifacts.downloadReport(notebook.id, status.artifactId);

// Generate slide deck
const slides = await client.artifacts.generateSlideDeck(notebook.id, {
  instructions: 'Create executive briefing...',
  format: 'presenter',
  length: 'default'
});

Changelog

0.2.3 (2026-03-21)

Bug Fixes

• Source import RPC params: Fixed importSources (IMPORT_RESEARCH_SOURCES / LBwxtb) parameter structure from [notebookId, [[url, title]]] to [null, [1], taskId, notebookId, [[null, null, [url, title], null*8, 2], ...]]. Previous format caused Google to return null — sources discovered during research were never imported.
• Source extraction double-nesting: Fixed parseResearchResult source parsing — sources at innerData[3] are wrapped as [[[url, title, snippet, 1], ...]] (double-nested). Parser was iterating the outer array and passing the entire source list as a single entry to parseResearchSource. Added unwrapping logic.
• taskId propagation: research() now passes taskId to importSources() — required by the IMPORT_RESEARCH_SOURCES RPC.

0.2.2 (2026-03-21)

Bug Fixes

• Research RPC params: Fixed parameter structure for START_FAST_RESEARCH and START_DEEP_RESEARCH — was using flat [notebookId, query, sourceType], corrected to [[query, sourceType], null, 1, notebookId] (fast) and [null, [1], [query, sourceType], 5, notebookId] (deep). Previous format caused Google to return null with error code [3].
• Poll RPC params: Fixed POLL_RESEARCH params from [notebookId, taskId] to [null, null, notebookId]. Poll returns all tasks for the notebook.
• Research result parsing: Rewrote parseResearchResult to handle actual response structure [[[taskId, innerData, ts, ts], ...]] instead of expected flat [statusCode, summary, sources]. Now correctly finds task by ID, extracts sources/summary, and detects completion status.
• Deep research timeout: Increased default from 120s to 900s (15 min) — deep research takes 5-15 minutes. Fast research keeps 120s default.
• CLI --timeout option: Added -t, --timeout <ms> flag to research web command for custom timeout values.
• Version string: Fixed hardcoded 0.1.0 in CLI to match package version.

RPC Parameter Reference

Method	ID	Params
START_FAST_RESEARCH	Ljjv0c	[[query, sourceType], null, 1, notebookId]
START_DEEP_RESEARCH	QA9ei	[null, [1], [query, sourceType], 5, notebookId]
POLL_RESEARCH	e3bVqc	[null, null, notebookId]

0.2.0 (2026-03-20)

Fork of notebooklm@0.1.1 by kaelen with extensive bug fixes and new features.

Bug Fixes

• Source status parsing: Sources always showed "processing" even when ready. The GET_NOTEBOOK response wraps status in [null, statusCode] array — added array traversal to extract the actual code.
• Source list RPC method: Changed from hizoJc (GET_SOURCE_FULLTEXT) to rLM1Ne (GET_NOTEBOOK) which actually returns sources.
• Source ID parsing: GET_NOTEBOOK wraps source IDs as ["uuid"] arrays — added unwrapping logic in parseSourceItem.
• Source list traversal: parseSourceList now correctly finds sources at the right nesting level in the GET_NOTEBOOK response.
• addText RPC params: Fixed parameter format from flat [notebookId, title, content] to correct nested array structure per Google's batchexecute protocol.
• addUrl RPC params: Fixed parameter format for both regular URLs and YouTube URLs.
• Source delete params: Fixed to use triple-nested sourceId [[[sourceId]]].
• Generation status codes: Fixed mapping — was 0=Pending, 1=InProgress, 2=Completed, 3=Error, should be 1=Processing, 2=Pending, 3=Completed.
• Artifact status parsing: Status is at item[4] in the LIST_ARTIFACTS response, not item[3].
• Artifact listing: Added required source-path=/notebook/{id} URL parameter to rpcCall — without it, LIST_ARTIFACTS returns null.
• getStatus method: Replaced fake GET_GENERATION_STATUS RPC (doesn't exist) with LIST_ARTIFACTS + find by ID.
• CREATE_ARTIFACT method ID: Changed from xpWGLf to R7cb6c.
• generateReport: Complete rewrite — fetches source IDs from notebook and builds correct nested params with source ID references.
• downloadReport: Now reads artifact.content from LIST_ARTIFACTS response instead of returning empty string.

New Features

• Slide deck generation: Added generateSlideDeck() API method and notebooklm generate slide-deck CLI command with --prompt, --format (detailed/presenter), and --length (default/short) options.
• Artifact download: Added notebooklm generate download <notebookId> <artifactId> <outputPath> CLI command for downloading artifact content to files.
• rpcCall source-path: Added optional sourcePath parameter to rpcCall for methods that require notebook context.

RPC Method Reference

Method	ID	Notes
GET_NOTEBOOK	rLM1Ne	Returns notebook + sources
ADD_SOURCE	izAoDd	URL, text, YouTube
DELETE_SOURCE	tGMBJ	Triple-nested source ID
LIST_ARTIFACTS	gArtLc	Requires source-path
CREATE_ARTIFACT	R7cb6c	All artifact types
DELETE_ARTIFACT	V5N4be
EXPORT_ARTIFACT	Krh3pd	To Docs/Sheets
IMPORT_RESEARCH_SOURCES	LBwxtb	Requires taskId

License

MIT