Local & self-hosted appslfnovo (open-source community)Open source
Open Notebook
Decision matrix
Open Notebook app decision matrix
Overview
Verdict: the best open-source stand-in for Google's NotebookLM if your sources must stay on your own machine. It chats over your PDFs, pages and videos and turns them into podcasts, with the model of your choice. Expect a Docker setup and basic citations; for a shared team knowledge base, AnythingLLM is easier.
Other good fits
Turning a reading list into an audio summary with one to four speakers, for a journal club or a team briefing.
Anyone who wants to pick the model: OpenAI, Anthropic, Google, or a local model through Ollama or LM Studio.
+ Pros
Read pros
Open Notebook is an open-source (MIT) research notebook, written as a private alternative to Google's NotebookLM. You collect sources in a notebook — PDFs, Office files, web pages, YouTube videos, audio — then chat with them, run summaries and extractions, and generate podcasts. The current release is 1.14 (July 2026).
Pick it when the sources can't leave your control. It runs in Docker on your computer or server. With a local model through Ollama, nothing goes to a cloud service.
Pick it when you want to choose the model. It supports more than 18 providers, including OpenAI, Anthropic, Google, Ollama and LM Studio, and any OpenAI-compatible endpoint.
Pick it when audio helps. The podcast generator writes and voices an episode from your notes, with one to four speakers.
− Cons
Pick something else when you want zero setup. Google's NotebookLM runs in the browser with a Google account.
Pick something else when a team needs shared workspaces and user accounts.AnythingLLM and Onyx are built for teams. Open Notebook protects an instance with one optional password.
Citations are basic. The project's own comparison says so. Check each answer against the passage it came from.
Small local models give weaker answers. Test on questions you can check before you trust it with new ones.
Avoid if
You need precise, sentence-level citations. The project itself rates its citations as basic.
Workflow
Create a notebook -> add approved sources -> configure a provider without exposing keys -> ask a limited question -> compare the output with the sources -> retain the source links and correct mistakes manually.
Cost
Read cost
Prices checked 23 September 2026: free software (MIT licence); you pay for the machine and for the model.
Machine: any computer or small server that runs Docker. For local models, the project's quick start lists small Ollama models needing 2–8 GB of GPU memory; larger, better models need more.
Model: with a cloud provider you pay its per-token rates. The project's OpenAI quick start estimates $1–5 a month for light use and $20–50 for heavy use.
Effort: updates, backups of the database volume, and security are yours. Change the default database credentials (root/root) and the encryption key before you store anything real, never publish the database port, and use the optional password if others can reach it.
Restrictions
The reviewed primary source is the evolving upstream repository. Its current README should be checked for supported providers, installation steps, licence, and deployment requirements before relying on an integration.
First project
Documentation-based suggestion — not run here.
Read first project
Lab: five papers for next week's journal club. Company: three supplier contracts and their price sheets.
Start Open Notebook with the project's Docker Compose file. Connect one provider under Models: an API key, or Ollama for fully local use.
Create a notebook and add the five documents.
Ask three questions you already know the answer to, such as "Which paper uses the smallest sample?" Open the referenced source for each answer.
If the answers hold up, ask the questions you actually have, then generate a 10-minute two-speaker podcast for the group.
You're done when every answer you keep matches a passage you found in the source yourself. T04-L01 teaches this check; T04-L02 extends it to a team.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Based on the Open Notebook repository (README, quick starts, Docker Compose docs, v1.14.0 release notes) and project website, checked 23 September 2026. We did not install it for this review; the worked example is an illustration.
Open Notebook repository · Sep 23, 2026 — MIT licence; 18+ providers; 1-4 speaker podcasts; content types; comparison with NotebookLM rating its own citations as basic.
Open Notebook v1.14.0 release · Sep 23, 2026 — Current release, July 2026, with provider and security hardening changes.
Local quick start · Sep 23, 2026 — Fully local setup with Ollama; small models need 2-8 GB of GPU memory; never publish the database port.
OpenAI quick start · Sep 23, 2026 — Cost estimate of $1-5/month light use and $20-50/month heavy use; replace the default encryption key.
Docker Compose installation · Sep 23, 2026 — Default database credentials root:root and how to change them.
Collect sources and search them in one workspace.
Produce assisted notes or audio drafts from collected material.
Choose local or cloud model providers for processing.
View sourced imageOfficial Open Notebook documentation screenshot.
lfnovo/open-notebook
MIT-licensed repository documentation; copyright and full permission notice retained.
MIT License
Copyright (c) 2024 Luis Novo
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE. Converted on 15 October 2026 to a WebP (quality 80) for faster page loads; the original image remains in the media library.
