Open Notebook: Self-Hosted Notebook LM Alternative
/open-notebook
Open Notebook is a self-hosted stand-in for Google's Notebook LM that trades polish for control over your data and model bill. Citation handling still lags — the README calls it 'basic references (will improve)' — but provider flexibility is real: point it at GPT-4, Claude, or a free local Ollama model without touching code. If you can't put source documents on someone else's cloud, that flexibility outweighs the rough edges.
Why you need a private AI research tool
Feeding proprietary contracts, medical notes, or unreleased research into a cloud AI notebook means trusting Google (or a startup) with data you can't take back. Notebook LM only runs on Google's cloud and only talks to Google's models, so there's no way to point it at a local Ollama instance or swap in a cheaper provider for a low-stakes task. Open Notebook exists for the case where source material is sensitive enough that it has to stay on hardware you control, but you still want the note-taking, citation, and podcast-generation workflow Notebook LM popularized.
What Open Notebook is
Open Notebook is an open-source, self-hosted implementation of Notebook LM's workflow: create notebooks, drop in PDFs, videos, audio, web pages, and Office docs as sources, then chat with, search, and generate podcasts from that material. It runs as a Docker Compose stack — Next.js/React frontend, Python/FastAPI backend, SurrealDB database — and talks to whichever AI provider you configure, from OpenAI and Anthropic to a local Ollama model.
Core features and supported AI providers
- ✓18+ configurable AI providers (OpenAI, Anthropic, Google, Groq, Ollama, LM Studio, Mistral, DeepSeek, and more) through the Esperanto library, swappable per task
- ✓Multi-modal source ingestion — PDFs, videos, audio, web pages, and Office documents land in the same notebook
- ✓Podcast generation with 1-4 speakers and custom Episode Profiles, versus Notebook LM's fixed two-speaker format
- ✓Full-text and vector search across every source in a notebook
- ✓Content Transformations — customizable actions that summarize or extract insights from a source on demand
- ✓A full REST API for scripting notebooks, sources, and chats outside the web UI
- ✓Reasoning-model support for DeepSeek-R1 and Qwen3, plus MCP integration for Claude Desktop and VS Code
- ✓Optional password protection for deployments exposed beyond localhost
What you can do with Open Notebook
- •Drop a stack of PDFs and web articles into a notebook, then chat with the combined source material instead of re-reading each file
- •Turn a research notebook into a multi-speaker podcast episode with a custom Episode Profile
- •Run entirely offline against a local Ollama or LM Studio model when source material can't leave the machine
- •Call the REST API from a script to add sources or pull notes programmatically instead of clicking through the UI
- •Connect Open Notebook to Claude Desktop or VS Code over MCP to query notebook contents from inside another tool
Installing with Docker in 2 minutes
Open Notebook ships a docker-compose.yml with two services: SurrealDB and the open_notebook app image. Download it with `curl -o docker-compose.yml https://raw.githubusercontent.com/lfnovo/open-notebook/main/docker-compose.yml`, then edit the OPEN_NOTEBOOK_ENCRYPTION_KEY line to a real secret — this key encrypts the API keys you'll store in the database, and the default value ships as a placeholder. Run `docker compose up -d`, wait 15-20 seconds, and open http://localhost:8502 for the web UI (the REST API listens on port 5055). The SurrealDB service defaults to root:root credentials and binds only to 127.0.0.1 for local debugging; the README warns to set SURREAL_USER and SURREAL_PASSWORD in a .env file before exposing the instance to a network. From there, go to Models in the UI, add a provider configuration and API key, test the connection, and sync models.
Advantages over Google Notebook LM
- ✓Provider choice: 18+ AI backends including free local options like Ollama, versus Notebook LM's single Google-only model
- ✓Podcast flexibility: up to 4 speakers with custom Episode Profiles instead of a fixed 2-speaker format
- ✓A full REST API for automation, which Notebook LM doesn't offer at all
- ✓Self-hosted deployment means source documents never leave infrastructure you control
- ✓Open source and MIT-licensed, so the codebase itself can be modified or audited
Current limitations and known gaps
- △Citations are, in the README's own words, 'basic references (will improve)' — noticeably behind Notebook LM's sourced citation UI today
- △Running it means operating three services yourself (SurrealDB, the app container, and whichever AI provider you pick) instead of a zero-maintenance hosted product
- △The default docker-compose.yml ships SurrealDB with root:root credentials, which the README says must be changed before exposing the instance to a network
- △Real-time UI updates and async content processing are still on the roadmap, not shipped yet
Other open source AI notebook tools
Frequently asked questions
Open Notebook itself is free and open source under the MIT license, so there's no subscription for the software. You still pay for whichever AI provider you connect — OpenAI or Anthropic usage costs money, but pointing Open Notebook at a local Ollama model avoids API fees entirely.
Open Notebook supports Ollama and LM Studio as AI providers, so it can run completely offline without sending data to any external API. The README includes a docker-compose-ollama.yml example specifically for this fully local setup.
Open Notebook supports 18+ AI providers through the Esperanto library, including OpenAI, Anthropic, Google, Groq, Ollama, LM Studio, Mistral, DeepSeek, Cohere, and OpenRouter. Provider support varies by capability — some cover LLM chat only, while others add embeddings, speech-to-text, or text-to-speech.
Open Notebook is self-hosted and lets you pick from 18+ AI providers, while Google Notebook LM runs only in Google's cloud on Google's own models. Open Notebook also supports 1-4 podcast speakers and a full REST API, both of which Notebook LM lacks, though its source citations are less polished than Notebook LM's today.
Open Notebook ships a full REST API alongside its web interface, exposed on port 5055 in the default Docker setup. It lets you create notebooks, add sources, and manage content programmatically instead of clicking through the UI, and it's documented in the project's API reference.
Open Notebook notebooks accept PDFs, videos, audio files, web pages, and Office documents as sources. Everything you add becomes searchable and usable in chat, notes, and podcast generation within that notebook.
Who should try it — and who should skip
Try Open Notebook if you already run a homelab or a small Docker stack and want an AI notebook that never sends source material anywhere you didn't choose — researchers with confidential documents, and anyone who wants to mix a cheap or local model for bulk reading with a stronger one for synthesis. Skip it if you just want something that works today with zero maintenance: Notebook LM's hosted citations and zero-setup UI are still ahead of what a self-run Docker stack gives you out of the box.
Related repositories
Still deciding about open-notebook?
One click hands the question to an AI along with this page — see what it says about open-notebook.