cyberivy
Open NotebookNotebookLM AlternativeSelf-hosted AIAI Research ToolsDocument ChatOpen Source AIOllamaPrivacy

Open Notebook: an AI research notebook on your own server

November 21, 2026

Aufgeschlagenes Notizbuch mit goldener Spiralbindung und leeren linierten Seiten, daneben ein gelber Stift auf einer gelben FlΓ€che

Open Notebook is a self-hosted alternative to Google's NotebookLM: collect sources, query them with citations, generate podcasts. Version 1.15.1 from 4 November 2026 closes security gaps.

What this is about

Google's NotebookLM showed how useful a notebook can be that reads your own documents and answers questions about them with source citations. The catch for many teams: everything runs in Google's cloud, with Google's models. Open Notebook by developer Luis Novo (GitHub: lfnovo) addresses exactly that. The project is open source under the MIT licence, runs on your own computer or server and has about 42,000 stars on GitHub (as of 21 November 2026).

The current occasion is version 1.15.1 from 4 November 2026, a security update the developers recommend to all users.

What Open Notebook actually does

You create notebooks and fill them with sources: PDFs, web pages, videos, audio files and other formats. Open Notebook extracts the content, indexes it for full-text and vector search and then offers a chat whose answers, according to the project description, cite the sources, notes and insights used. On top of that come freely definable transformations, such as summaries following your own template, and a podcast feature with one to four speakers.

You are free in your choice of models: according to the README, more than 20 providers are supported, including OpenAI, Anthropic, Google, Mistral and Ollama, as well as OpenAI-compatible servers such as LM Studio or vLLM. With Ollama you can work entirely locally. Installation uses Docker Compose with two services: the SurrealDB database and the application itself with interface, programming interface (REST API) and background worker.

Since version 1.13.0 from July 2026, there are optional engines: Docling for text recognition in scanned PDFs and Crawl4AI for JavaScript-heavy web pages.

Why it matters

In research, journalism, consulting or legal departments, sources are often confidential. A self-hosted tool with a freely chosen model allows sensitive documents to stay in-house and costs to be steered through the model choice. The REST API also makes Open Notebook automatable, which NotebookLM does not offer according to the comparison in the README. Independent reviews, for example on andrew.ooo, classify the project as a serious alternative with noticeable setup effort.

In plain language

Open Notebook is like your own reserved shelf in a library, but with an assistant next to it. You put the books on the shelf yourself, the assistant reads them and, with every answer, points to the page it came from. The shelf stands in your own house, not at the publisher's.

A practical example

A small consultancy is preparing a study on heat pump subsidies. It loads 40 sources into a notebook: 25 PDF reports, 10 web pages and 5 recordings of expert talks. The model runs locally via Ollama so that client documents never leave the building. Open Notebook answers the question of which subsidy rates the sources mention with references to 6 documents. Two figures contradict each other; the consultant checks both originals and records the result as a note. Finally she generates a ten-minute podcast with two voices as an introduction for the team.

Scope and limits

  • Operations are your job: According to the README, the interface (port 8502) and API (port 5055) are reachable on all network interfaces by default and without a password. Anyone running it on a network must restrict the ports to localhost or set a password, and choose their own encryption key for stored API keys.
  • Security updates are mandatory: Version 1.15.1 closes gaps in source processing and hardens URL fetching against internal and cloud metadata addresses. The details are to be published as security advisories in the repository. Anyone who delays updates carries the risk.
  • Quality depends on the model: With small local models, answers and source attribution become weaker. According to the README, Anthropic and Groq offer no embedding models, so you then need a second provider. Citations do not replace checking the original.

A sensible first test: start Open Notebook via Docker Compose bound to localhost only, fill a notebook with five documents you know well and check whether the citations are correct for ten control questions.

SEO & GEO keywords

Open Notebook, lfnovo, NotebookLM alternative, self-hosted AI, AI research, document chat, Ollama, SurrealDB, AI podcast, open source, data privacy

πŸ’‘ In plain English

Open Notebook is a free program into which you load your own documents and then ask questions about them. The answers point to the sources. It runs on your own computer, and you choose which AI model works behind it.

Key Takeaways

  • β†’Open Notebook is an open-source, self-hosted alternative to Google's NotebookLM under the MIT licence.
  • β†’According to the README, more than 20 model providers are supported, and Ollama enables fully local operation.
  • β†’Sources such as PDFs, web pages, videos and audio can be searched, queried in a chat with citations and turned into podcasts.
  • β†’Version 1.15.1 from 4 November 2026 is a recommended security update.
  • β†’By default, the interface and API are reachable on the network without a password and must be secured.

FAQ

What does Open Notebook cost?

The software is free. Costs arise only for model calls at cloud providers, or for your own hardware if you use local models.

Do my documents really stay local?

The data sits on your own server. But if you use a cloud model, the relevant text excerpts are sent to that provider. Only with local models, for example via Ollama, does everything stay in-house.

How much effort is the installation?

You need Docker with Compose, a configuration file and your own encryption key. After that, model providers are set up in the interface.

Does Open Notebook fully replace NotebookLM?

Functionally it covers the core ideas and goes further with an API and free model choice. In return, you take on operation, updates and security yourself.

Sources & Context