notebook

Notebook Plugin for Claude Code

A NotebookLM-style research assistant plugin for Claude Code. Ingest PDFs and text files, then generate rich outputs from your sources — all from the command line.

What it does

1. Ingest PDFs or text files into a local vector store (with automatic OCR for scanned documents)
2. Query your sources using semantic search powered by local embeddings
3. Generate 8 different output types grounded in your source material
4. Review outputs through a 3-agent pipeline that checks accuracy against sources

Output types

Type	Generator	Output files
Flashcards	Interactive study cards with flip animations, tag filtering, keyboard nav	.json + .md + .html
Quiz	50-question assessment with live scoring, letter grades, answer review	.json + .md + .html
Report	Structured analysis with executive summary and evidence-backed findings	.md + .docx
Slide Deck	Professional presentation with speaker notes	.pptx
Mind Map	Interactive diagram with pan/zoom canvas and Mermaid rendering	.mmd + .html
Infographic	Visual summary with scroll animations, floating TOC, and stat callouts	.html
Data Table	Sortable/filterable table with search, pagination, and CSV/JSON export	.csv + .json + .md + .html
Audio Overview	Podcast-style two-host discussion using Kokoro-82M neural TTS	.wav

All HTML outputs feature a unified dark theme with Inter typography, Lucide icons, keyboard shortcuts, and responsive design.

Installation

/init

This registers the plugin skills and prepares the environment. You can also install manually:

From GitHub (marketplace)

/plugin marketplace add damionrashford/notebook-plugin
/plugin install notebook@notebook-marketplace

Direct install

/plugin install damionrashford/notebook-plugin

Requirements

• Bun 1.0+ (for scripts with npm dependencies — auto-installs on import)
• Node.js 18+ (for zero-dependency scripts and the dashboard server)
• uv (for Audio Overview — runs Kokoro TTS with inline script deps)
• macOS (required for Audio Overview AIFF→WAV transcode via afconvert)

Usage

Ingest sources

/notebook:ingest path/to/document.pdf
/notebook:ingest path/to/notes.txt

Supports text-based PDFs and scanned PDFs (auto-OCR via Tesseract.js). Multiple files can be ingested into the same store.

Generate outputs

/notebook:generate flashcards
/notebook:generate quiz
/notebook:generate report
/notebook:generate slide-deck
/notebook:generate mind-map
/notebook:generate infographic
/notebook:generate data-table
/notebook:generate audio-overview

Add an optional topic to focus the output:

/notebook:generate flashcards neural networks
/notebook:generate report transformer architecture

View dashboard

/notebook:dashboard

Generates a NotebookLM-style HTML dashboard showing your ingested sources, available output types, and generated files.

Direct CLI access

# Query the vector store directly
bun skills/ingest/scripts/query.mjs "your question" --top-k 15

# List ingested sources
node skills/ingest/scripts/list.mjs

Architecture

3-Agent pipeline

The plugin uses three orchestrated agents:

Agent	Role	Access
Researcher	Deep document analysis — runs 5-10 queries to explore sources from multiple angles	Read-only
Writer	Generates all output artifacts from research findings	Full write access
Critic	Reviews outputs against source material, catches errors and gaps	Read-only

Pipeline flow: Researcher analyzes sources -> Writer generates artifact -> Critic reviews against sources -> Writer revises if needed

RAG pipeline

• PDF parsing: pdf-parse with Tesseract.js OCR fallback for scanned documents
• Chunking: Semantic text splitting with configurable overlap
• Embeddings: Local model via @huggingface/transformers (all-MiniLM-L6-v2, 384 dimensions)
• Vector store: Vectra (local, file-based — no external services needed)
• Storage: ~/.notebook-plugin/stores/<project-hash>/ — persists across sessions

Key paths

Path	Purpose
~/.notebook-plugin/	Vector store and embedding model cache
./output/	All generated artifacts
~/.claude/plugins/data/notebook/	Installed npm dependencies (managed by hook)

3 Skills