If you're like me, you've probably got a YouTube watchlist spanning hundreds of videos. It's basically impossible to stay up to date with everything I want to see. I've tried using NotebookLM and accompanying extensions to try and stay on top of things, but, unlike my colleague, I haven't been able to make much headway. What I really wanted and needed was a self-hosted tool that worked independently in the background, processing a list of YouTube links, summarizing them, and allowing me to have a conversation with it for further insights. That sounds like I'm expecting a lot from a tool. Turns out, tl/dw does exactly that. The cheekily named tool is specifically designed for you to wrangle videos that were too long and that you didn't get around to watching. Here's why it might just be the best self-hosted option out there if you want an alternative to NotebookLM for summarizing YouTube videos.
Tl/dw: What's this app all about?
Fixing the YouTube overload problem
Cheeky as it sounds, the name basically tells you all you need to know about TL/DW. This open-source, self-hosted app can take in videos, audio files, articles, and even PDFs, then help you transcribe, summarize, and chat with them using the power of LLM models. Sounds like NotebookLM? Except this time, it runs entirely locally, giving you full control over your data.
The problem this app is tackling is simple and relevant to most users. There's just way too much content on YouTube. Between podcasts, documentaries, interviews, and explainers, I just don't have enough time to watch everything. Heck, if a video doesn't have timestamps, I'm very likely to bounce off. Tl/dw isn't the first AI video summarizer that I've come across, but most such apps lock you into their own ecosystem. Similarly, NotebookLM is great, but that'll lock you in again. Tl/dw flips that with its easy and straightforward Docker install and full data ownership. I'm all in for that.
So... what all can it do?
A full AI toolbox
What I like about tl/dw is that the functionality goes well beyond just YouTube summarization. And even that is made extremely simple with an easy ingestion process. You can drop in a YouTube link or just drop in a couple of URLs. You can even drop in a text file full of links or a YouTube playlist. There's a lot of flexibility here, and tl/dw makes it a cinch to catch up with a full series of playlists in one go.
Once you've dropped in your media, be it YouTube or perhaps a document, this self-hosted app handles transcription with its local Whisper model. Of course, there are options available to pick from, and you've got the choice to pick models based on your preference for speed or accuracy. From there, it goes through the standard process of chunking so that it can generate summaries without blowing through the context window.
Tl/dw lets you define whatever LLM model you want to use for the actual summarization process. This can be anything from OpenAI to HuggingFace, Anthropic, Gemini, and many others. Of course, the real incentive here is to run your local model using Ollama if you want to stay completely offline.
With all that configured, it is pretty straightforward to start getting summarizations for whatever you drop into the app. Now, summarization is just one part of the experience. The app supports RAG, or Retrieval Augmented Generation, which allows you to ask a question and have it pull up relevant context from your own media library before responding. Pretty cool.
There's a built-in chat experience too, which lets you ask questions in plain English, and it will use your imported videos and documents to answer queries in natural language. Depending on your use case, the app can also be used to compare the response of different LLM APIs to see which one gives you the most accurate or most appropriate results.
If that wasn't enough, TL/DW takes an all-encompassing approach to LLMs. Switching over between the tabs reveals a few additional features, like a built-in grammar and tone check utility, the ability to generate prompts, or create a database of tools. Honestly, the app is a pretty full-featured toolbox of AI features with very few direct alternatives. Being self-hosted, of course, all your data stays on your server. But that goes without saying.
Why I'm switching over to tl/dw
The biggest reason why I'm seriously considering dropping NotebookLM and switching over to tl/dw is the fact that it gives me total ownership of how I want to work around my data. I can choose everything from the AI model I want to use, including local models, to how I want to use them. There's full ownership over data, and it surfaces a whole range of utilities that go well beyond its core feature of summarization. But even when it comes to its core utility, the app does a fantastic job of turning YouTube videos and media into something you can understand and grasp in a limited amount of time. For me, this is the best and most full-featured NotebookLM alternative I've come across.
