I Found the Self-Hosted NotebookLM Devs Actually Want (Open-Notebook)

BBetter Stack
컴퓨터/소프트웨어창업/스타트업AI/미래기술

Transcript

00:00:00Notebook LM, it's amazing.
00:00:02You upload a paper, a document, maybe even a code base,
00:00:05and now you can chat with it and summarize it.
00:00:08But then there's the part nobody likes talking about.
00:00:10You still have to upload all that stuff to Google.
00:00:14This is Open Notebook.
00:00:15It has over 27,000 stars on GitHub.
00:00:18It's self-hosted, and it asks a very simple question.
00:00:22What if you could get the Notebook LM experience,
00:00:24but with dev-level control?
00:00:26Today, I'm gonna spin it up, test the workflow,
00:00:29compare it to Notebook LM and anything LLM,
00:00:32and answer the real question.
00:00:34Is this actually useful for devs,
00:00:36or is it just another AI wrapper?
00:00:43Open Notebook is a privacy-first,
00:00:45self-hosted alternative to Notebook LM,
00:00:47but that undersells it a little bit,
00:00:49because this is not just Notebook LM, but open source.
00:00:53It gives you a Notebook LM-style research workspace,
00:00:57multi-model support, podcast generation, local first options,
00:01:01and a REST API you can actually build on top of.
00:01:04And that's the part a lot of people are actually caring about.
00:01:07Most people look at this and they think one thing.
00:01:09Cool, I can make an AI podcast from PDFs.
00:01:12Bravo.
00:01:14Devs look at it and we think something slightly different.
00:01:16Can I plug this into my workflow?
00:01:19That's one real question.
00:01:20Can I use local models with it?
00:01:22Can I automate research summaries?
00:01:25And answering all those questions
00:01:26is where Open Notebook gets interesting.
00:01:28You're not locked into Gemini.
00:01:30You can use different providers,
00:01:32including local models through Alama.
00:01:34You can self-host it.
00:01:35You can customize the podcast experience
00:01:37with different speaker profiles.
00:01:39And because there's an API,
00:01:41this can become part of your stack.
00:01:43Not just another tab in your browser.
00:01:45If you enjoy coding tools that speed up your workflow,
00:01:48be sure to subscribe.
00:01:49We have videos coming out all the time.
00:01:51All right, let's actually run this
00:01:52so you can see it in action.
00:01:55Open Notebook is Docker first.
00:01:57That's awesome.
00:01:58So if you're already comfortable with containers,
00:02:00this is really familiar territory.
00:02:02Run the Compost setup,
00:02:04wait for the services to spin up,
00:02:06and then open the app in your browser.
00:02:08Now that it's going, we can create a new notebook.
00:02:10Think of a notebook like a project-specific research workspace.
00:02:13Instead of dumping everything into one giant AI chat,
00:02:17you can separate things out.
00:02:19One notebook for code bases,
00:02:21one for project research,
00:02:22one for academic papers,
00:02:24internal docs,
00:02:25all that stuff.
00:02:26Now we can add in our sources.
00:02:29This could be things like a PDF,
00:02:31a readme,
00:02:31documentation,
00:02:32a research paper,
00:02:33or really anything you want the system
00:02:35to actually reason over.
00:02:37And the important word there is source,
00:02:39because the goal is not just generic AI chat.
00:02:43The goal is grounded answers based on the material that you give it.
00:02:48So let's ask now a more technical question here.
00:02:51Maybe I can ask something along the lines of,
00:02:53what are the main components of this project,
00:02:55and what would I need to change if I want to extend it?
00:02:58It's doing its thing.
00:03:00This is the basic notebook LM style experience.
00:03:03We add sources,
00:03:04you ask questions,
00:03:04you get answers grounded on those sources.
00:03:07But here's the main thing.
00:03:08This is where it stops feeling like a clone of something
00:03:11and starts feeling more dev-friendly,
00:03:13more of something maybe on its own.
00:03:15You can choose which model provider you want to use,
00:03:18and we're not locked into that vendor like Gemini,
00:03:20like I said.
00:03:21Hosted models,
00:03:22absolutely fine.
00:03:24Local models,
00:03:25also absolutely fine.
00:03:27You get to pick the trade-off
00:03:28between quality,
00:03:29speed,
00:03:30cost,
00:03:30and privacy.
00:03:31And then there's the feature
00:03:32most of us will click first,
00:03:34the podcast generator.
00:03:36Notebook LM made AI podcasts
00:03:38feel actually pretty cool.
00:03:40If you haven't played with it,
00:03:41maybe you should.
00:03:42If I run it here,
00:03:44well,
00:03:44something else happens.
00:03:46Take a listen.
00:03:46It's a game changer for researchers
00:03:48looking for autonomy and privacy.
00:03:50Absolutely, Alex.
00:03:52I think one of the coolest aspects of Olama.
00:03:54Cool, right?
00:03:55But Open Notebook gives you more control
00:03:57over that format.
00:03:58You can generate podcasts
00:04:00from your sources,
00:04:01configure the structure,
00:04:02and use multiple speaker profiles
00:04:04instead of being stuck
00:04:05with one fixed style.
00:04:07So instead of getting
00:04:08generic AI hosts
00:04:10explain a PDF,
00:04:11you can create something
00:04:12more specific.
00:04:14You could say something like
00:04:15a product manager
00:04:16and a backend developer
00:04:17debating an architecture doc.
00:04:19And that sounds small
00:04:20until you use it on something
00:04:22that's honestly painful.
00:04:24A long RFC,
00:04:25a dense white paper,
00:04:26a boring API spec,
00:04:28all that stuff.
00:04:29It's a way to make dry information
00:04:30just easier to consume.
00:04:32Now, let's compare it
00:04:33to the tools we already know
00:04:35that are out there.
00:04:36Let's start with Google Notebook LM.
00:04:39Notebook LM is great.
00:04:40It's easy.
00:04:41It works really well.
00:04:41And for a lot of us,
00:04:43honestly, that's probably enough.
00:04:45But the trade-off behind all that
00:04:46is the control.
00:04:48Open Notebook gives you
00:04:49self-hosting,
00:04:50multi-model support,
00:04:51local model options,
00:04:52customizable podcasts,
00:04:54and API access.
00:04:55So if you're working
00:04:56with sensitive docs,
00:04:58private research,
00:04:59or internal material,
00:05:01Open Notebook has
00:05:01the stronger privacy story here.
00:05:04Now, here's the catch to all that.
00:05:06Is Open Notebook as smooth
00:05:07as Notebook LM?
00:05:08No, not always.
00:05:10Notebook LM has the advantage
00:05:12of being polish-hosted product.
00:05:15It's from Google, right?
00:05:16Open Notebook is more flexible,
00:05:17but it still feels like
00:05:18a dev-oriented open-source project.
00:05:20That's not a deal-breaker.
00:05:22It just means you should know
00:05:23what you're actually choosing.
00:05:24Now, compare it to anything LLM.
00:05:27Honestly, that is a really cool tool.
00:05:30Anything LLM is also popular
00:05:31in the self-hosted AI space,
00:05:33but it wins in a different way.
00:05:35It's easier to get started with.
00:05:37It has a desktop app.
00:05:38It has no code agent workflows.
00:05:40That's great.
00:05:41For a non-technical user,
00:05:42that may be the first step.
00:05:44But Open Notebook feels more focused
00:05:46on the Notebook LM-style
00:05:48research experience.
00:05:50Now, let's be honest
00:05:50about what people like
00:05:51and what people are actually
00:05:52complaining about.
00:05:53The huge win here
00:05:54is going to be privacy.
00:05:56If your work involves sensitive docs,
00:05:58private code, client research,
00:06:00or anything you would be hesitant
00:06:02to upload to a hosted AI tool,
00:06:04then self-hosting really matters here.
00:06:06That is the main reason
00:06:07that Open Notebook
00:06:08actually exists in the first place.
00:06:10Then you have model flexibility.
00:06:12You're not forced into one provider.
00:06:14Yes, I can choose the ones I want.
00:06:16Huge win.
00:06:17That means we can choose
00:06:19what we need
00:06:19based on what we're working with,
00:06:21but it also creates a new problem.
00:06:24You have to make a choice.
00:06:25We also have the podcast customization.
00:06:28Tried on a huge project spec,
00:06:29a dense API, Docker,
00:06:31a long research paper,
00:06:32and it actually starts to make sense.
00:06:33Finally, the API is a big deal
00:06:35for a lot of us.
00:06:36You can imagine workflows
00:06:38like creating research briefings
00:06:40from GitHub issues
00:06:41or sending outputs
00:06:43into Slack, Linear, or Notion.
00:06:45Great.
00:06:46Now, the bad thing about all this,
00:06:48or maybe things
00:06:49that we don't quite like first,
00:06:50setup is Docker first.
00:06:52For most of us, honestly, that's fine.
00:06:54For everyone else,
00:06:55maybe that's a wall.
00:06:56This is not yet the download one app
00:06:59and everything just works.
00:07:00Second, it's still a newer project,
00:07:03so some things are still catching up.
00:07:05And then quality depends
00:07:06on your models and your setup.
00:07:08So the honest take here
00:07:10is kind of simple.
00:07:11Open Notebook is not perfect.
00:07:12Then again, no tool is perfect.
00:07:14That's why we have
00:07:14all these different tools.
00:07:16But the direction it's going in
00:07:17is very good.
00:07:18It's not for everyone,
00:07:19but give it a try
00:07:20if you want self-hosted
00:07:21research backend,
00:07:23if you have docs
00:07:23you don't want to just upload to Google,
00:07:25or if you want to build
00:07:27custom workflows on top of the API.
00:07:30The stack includes
00:07:31modern front-end,
00:07:32a Python backend,
00:07:33SurrealDB,
00:07:34and an AI abstraction layer
00:07:35designed to work across providers.
00:07:37So it can feel like something
00:07:39you can actually extend,
00:07:41not just something that we use.
00:07:42If you enjoy coding tools like this,
00:07:44be sure to subscribe
00:07:45to the Better Stack channel.
00:07:46We'll see you in another video.

Key Takeaway

Open Notebook provides a privacy-first, developer-oriented research environment that enables local model control, custom podcast generation, and API integration, offering a flexible alternative to Google's hosted NotebookLM.

Highlights

  • Open Notebook is an open-source, self-hosted alternative to Google's NotebookLM that holds over 27,000 stars on GitHub.

  • The platform supports multi-model integration, allowing users to choose between various hosted providers or local models via Ollama to balance privacy, speed, and cost.

  • Developers can leverage the built-in REST API to integrate research summaries directly into external workflows like Slack, Linear, or Notion.

  • Users can generate customizable AI podcasts from uploaded documents by configuring speaker profiles and structures instead of relying on a single, fixed style.

  • The application relies on a Docker-based deployment, featuring a stack comprised of a modern front-end, Python backend, and SurrealDB.

  • Grounding answers in specific user-provided sources—such as PDFs, readmes, and API specs—is the primary design goal, rather than generic AI interaction.

Timeline

Open Notebook Overview and Capabilities

  • Open Notebook acts as a self-hosted, open-source replacement for the research workspace functionality found in NotebookLM.
  • The tool provides multi-model support, podcast generation, and a REST API for programmatic access.
  • This platform aims to shift the user from passive consumption to active integration within a development stack.

While NotebookLM allows users to upload documents for summarization and chat, it requires sending data to Google. Open Notebook addresses this by offering a self-hosted environment with dev-level control. It supports various LLM providers, including local options, and includes an API for building automated research workflows.

Workflow Integration and Deployment

  • Deployment utilizes Docker, requiring users to run a Compose setup.
  • Notebooks function as project-specific research workspaces, allowing for the separation of code bases, papers, and internal documentation.
  • Users can select between hosted and local models to optimize for privacy, speed, and cost tradeoffs.

Setting up the environment involves standard Docker container management. Once active, users define distinct notebooks for different projects, preventing the issues associated with dumping all data into a single AI chat. The system grounds its answers exclusively in the sources provided by the user.

Podcast Customization and Tool Comparison

  • The podcast generator allows configuration of speaker profiles to discuss specific documents.
  • Open Notebook offers more structural control than the fixed styles found in Google's product.
  • Comparison with AnythingLLM reveals that while AnythingLLM focuses on ease-of-use and no-code agents, Open Notebook emphasizes the research workspace experience.

The podcast feature transforms dry information like white papers or API specifications into consumable audio formats. Unlike NotebookLM, which provides a polished but opaque experience, Open Notebook prioritizes user control and privacy. It serves as a middle ground between high-polish hosted products and simpler, desktop-focused AI tools.

Advantages, Constraints, and Technical Architecture

  • Privacy is the primary benefit, as it enables the analysis of sensitive documents and private code without third-party uploads.
  • The API allows for automated pipelines, such as creating briefings from GitHub issues.
  • Docker-based setup and the project's relative youth represent the main barriers to entry.

The platform is built with a modern front-end, Python backend, and SurrealDB, providing an extensible architecture for developers. While it lacks the immediate polish of a major tech corporation's product, it provides the necessary infrastructure for users who require custom workflows and data sovereignty.

Community Posts

No posts yet. Be the first to write about this video!

Write about this video