Your Research, Your Hardware, Your Rules.

Open Notebook is the self-hosted alternative — 18+ AI providers, no source caps, your data stays on your hardware. Docker setup in 2 minutes.

Deploy in 2 Minutes Save 20+ Hours

What Is Open Notebook

It's a full NotebookLM-style research workspace you run yourself — ingest PDFs, URLs, YouTube, audio, chat with documents, generate multi-speaker podcasts — using any AI model you choose instead of being locked to Gemini.

Built by lfnovo, 31K+ GitHub stars, MIT licensed. The key difference from NotebookLM: a REST API on port 5055, editable transformation prompts, 1-4 speaker podcast profiles, and per-task model routing (cheap model for summaries, Claude for analysis, local Ollama for private docs).

Who this is for: Developers who already use NotebookLM but need more — more sources, more privacy, more model choice, more automation via API. Not a NotebookLM tutorial — a deployment and production guide.

Where to Start

Deployment

Docker → Ollama → nginx + TLS → backup → multi-user. Copy-paste ready, 5 steps.

Hardware Sizing

From $5 VPS to RTX 4090. Three tiers with VRAM math. Stop guessing what GPU you need.

Model Guide

Best model combos for quality, speed, and cost. GPT-4o vs Claude vs local Ollama — benchmarked.

vs NotebookLM

Honest 10-dimension comparison. Where Open Notebook wins. Where it doesn't.

FAQ

Ollama OOM errors, embedding bugs, SurrealDB permissions, model selection — solved.

Production Manual

Full deployment configs, 30+ error fixes, monitoring, CI/CD. PDF + Markdown, $19.

Why Self-Host Your Document AI

1. Your data never leaves your network

With Ollama, Open Notebook processes everything on your hardware. For legal research, proprietary code, healthcare documents — this is a hard requirement. NotebookLM stores everything on Google's servers.

2. 18+ AI providers. No vendor lock-in.

OpenAI, Anthropic, Gemini, Mistral, DeepSeek, Ollama, LM Studio, Groq, xAI, and more — assign different models to different tasks. NotebookLM gives you Gemini. That's it.

3. Cost flips at scale

NotebookLM Plus costs ~$20/month/user. A 20-person team pays $400/month with Google seeing everything. Open Notebook with cloud APIs: $10-30/month total. With Ollama: $0 (just electricity).

Open Notebook vs NotebookLM

Dimension	Open Notebook	Google NotebookLM
Privacy	100% local possible	Data on Google servers
Setup time	2 min Docker → working	0 min — sign in, start
AI models	18+ providers	Gemini only
Source limits	No hard caps	50 sources, 500K words
API access	Full REST API	None
Cost at scale	$0-30/month	~$20/user/month
Citations	Basic references	Exact passages, highlights
Offline	Full offline (Ollama)	Internet required

Full 10-dimension comparison →

Quick Start

mkdir open-notebook && cd open-notebook
curl -o docker-compose.yml https://raw.githubusercontent.com/lfnovo/open-notebook/main/docker-compose.yml
sed -i 's/change-me-to-a-secret-string/YOUR-ENCRYPTION-KEY/' docker-compose.yml
docker compose up -d

Open http://localhost:8502 — working in 2 minutes. Full deployment guide →

Common gotcha: If using Ollama inside Docker, the URL must be http://ollama:11434 not http://localhost:11434. Localhost inside the container points to itself, not the Ollama container. Details in Step 2.

FAQ

What's the cheapest way to get started without a GPU?

Use Groq (free tier) or OpenAI API. Open Notebook needs only 4 GB RAM — a $5/month DigitalOcean droplet runs it fine with cloud APIs.

Which model combination gives the best balance of quality and cost?

Chat/Analysis: GPT-4o-mini or Claude Sonnet. Embeddings: text-embedding-3-small or qwen3-embedding (free, local). TTS: ElevenLabs. Zero-cost mode: Groq free tier + Ollama embeddings. Full model guide →

Is Open Notebook production-ready?

Yes, with the right setup. It's running in production at multiple companies for private document research. The Docker setup, nginx reverse proxy, and backup procedures in our Production Manual cover what you need. The main gaps are citations (not as good as NotebookLM) and multi-user isolation (still basic — use separate instances per team for now).

How do I keep my data safe when upgrading?

Your data lives in Docker volumes — not inside containers. docker compose pull && docker compose up -d upgrades safely. Always take a backup before major version bumps. Backup script here.

The free guides get you running. The Production Manual keeps you running — 100+ pages, full deploy configs, 30+ error fixes, monitoring, CI/CD. Save 20+ hours of debugging.

Get the Production Manual — $19