open-notebook: The Open-Source Notebook LM Alternative That Supports 15+ AI Providers — Self-Hosted, 28,000 Stars — Setup Guide 2026

Odysseus: The Self-Hosted AI Workspace That Hit 63 • Odysseus: Self-Hosted AI Workspace with 10+ Built-in Tools open-notebook — your self-hosted RAG knowledge base with multimodal audio

Introduction #

Google NotebookLM jumped to 1 million weekly active users within months of launch, proving that everyone needs a personal AI research assistant. But what if your documents are sensitive? What if you want to run it on your own infrastructure? open-notebook (28,200 GitHub stars) is the open-source answer — a self-hosted RAG knowledge base that ingests documents, answers questions with citations, and generates AI-powered audio “podcast” episodes from your sources. Unlike NotebookLM, it supports 15+ AI providers including Claude, GPT-4, local models via Ollama, and OpenRouter. In an era where document AI is critical but privacy matters, open-notebook gives you both.

What Is open-notebook? #

open-notebook is a self-hosted RAG (Retrieval-Augmented Generation) knowledge base that transforms your documents into an interactive AI-powered research workspace. Think of it as the intersection between a document question-answering system and an AI podcast generator.

Key capabilities:

• Document ingestion — Upload PDFs, markdown, text files, URLs, and more
• RAG-based Q&A — Ask questions about your documents; get answers with source citations
• Audio episodes — Generate AI-powered audio summaries that sound like a conversation between two hosts
• 15+ AI providers — Claude, GPT-4, Gemini, local models via Ollama/vLLM, OpenRouter, and more
• Self-hosted — Run on your own server, your own GPU, your own privacy

The project is built with Next.js (frontend) and Python FastAPI (backend). It uses vector databases for document embedding and retrieval, with a modern web interface for document management and conversation.

How open-notebook Works #

open-notebook operates through a three-stage pipeline:

Stage 1: Document Ingestion #

Raw Documents → Chunking → Embedding → Vector Storage

1. Upload — Import documents in multiple formats (PDF, MD, TXT, DOCX, URL)
2. Chunking — Split documents into semantic chunks using configurable strategies
3. Embedding — Generate vector embeddings for each chunk using a configured AI provider
4. Storage — Store embeddings in a vector database (Qdrant, Weaviate, or Supabase/pgvector)

Stage 2: Question Answering #

User Question → Embedding → Vector Search → Context Assembly → LLM Response

1. Query — User asks a question about their documents
2. Embedding — The question is embedded using the same model
3. Search — Vector similarity search finds the most relevant document chunks
4. Context assembly — Relevant chunks are assembled into a prompt context
5. LLM response — The configured AI provider generates an answer with source citations

Stage 3: Audio Episode Generation #

Documents → Script Generation → Multi-host TTS → Audio Episode

1. Document analysis — The system analyzes connected documents to identify key topics
2. Script generation — An LLM generates a dialogue script between two “hosts”
3. TTS synthesis — Text-to-speech converts each host’s lines into audio
4. Episode assembly — Audio clips are stitched together into a polished episode

┌──────────────────────────────────────────────────┐
│              open-notebook UI                     │
│  ┌──────────┐  ┌──────────┐  ┌──────────────┐   │
│  │ Documents│  │  Chat    │  │  Audio Episodes│   │
│  │  Manager │  │ Interface│  │  (Podcast Mode)│   │
│  └──────────┘  └──────────┘  └──────────────┘   │
├──────────────────────────────────────────────────┤
│         RAG Pipeline (Ingestion + QA)             │
├──────────────────────────────────────────────────┤
│   Vector DB (Qdrant / Weaviate / pgvector)        │
├──────────────────────────────────────────────────┤
│  AI Providers: Claude | GPT-4 | Ollama | OpenRouter│
└──────────────────────────────────────────────────┘

open-notebook architecture: three pipelines, one unified interface

Installation & Setup #

Docker Compose (Recommended) #

curl -o docker-compose.yml https://raw.githubusercontent.com/lfnovo/open-notebook/main/docker-compose.yml
docker compose up -d

Access the UI at http://localhost:3000.

Environment Configuration #

Edit the .env file with your API keys and provider configuration:

• Minimum: one of ANTHROPIC_API_KEY, OPENAI_API_KEY, or OLLAMA_HOST

Self-Hosted with GPU Acceleration #

For faster embedding and generation, run with GPU support:

# Ollama with GPU
curl -fsSL https://ollama.com/install.sh | sh
ollama pull nomic-embed-text:latest
ollama pull llama3.2:3b

# open-notebook with Ollama backend
# In .env:
# AI_PROVIDER=ollama
# OLLAMA_HOST=http://localhost:11434
# EMBEDDING_MODEL=nomic-embed-text
# COMPLETION_MODEL=llama3.2:3b

docker compose up -d

Importing Existing Notes #

Open-notebook can import from popular note formats:

# Import Obsidian vault
open-notebook import --source obsidian --path /path/to/vault

# Import Notion database
open-notebook import --source notion --api-key $NOTION_API_KEY --db-id $NOTION_DB_ID

# Import plain markdown files
open-notebook import --source markdown --path /path/to/md-files

# Import PDFs for RAG (extracts text + generates embeddings)
open-notebook import --source pdf --path /path/to/papers/*.pdf

Integration with 15+ AI Providers #

open-notebook supports a wide range of AI providers through a unified configuration interface:

Supported Providers #

| Provider | Type | Embedding | Chat | Audio | Cost | |