Cruxly's AI Notebook: How It Works and How It Differs From NotebookLM

Six months ago Cruxly had three separate tools: a YouTube video analyzer, a PDF analyzer, and a document chat, each with its own page and its own way of starting. Now it's one notebook: paste a link to a video, upload a PDF, or drop a link to an article, and it all lands in one place with one shared chat. That's not a rebrand, it's a different way of working: not "analyze one file and close the tab," but "pull together several sources on one task and talk to them as a whole." Google has been building exactly that idea into NotebookLM for years, and the comparison is unavoidable once two products solve the same problem. This guide covers how Cruxly's notebook works technically, how it compares to NotebookLM, and when that difference actually matters.

What an AI notebook actually is

Notebook, workspace, whatever the name, different products call the same idea different things: a container where you pull together several sources on one topic or task instead of analyzing them one at a time in isolation. On Cruxly a source can be a YouTube video, a PDF document, or a link to any web page. Each source gets its own structured breakdown: a TL;DR, a summary, key points. But the real reason to keep them together rather than analyzing each separately is the shared chat: it sees every source in the notebook at once and can answer a question that needs two documents compared, not one read in isolation.

A short practical guide: building a notebook from several sources

Why this is one tool, not three

Cruxly's video analysis, PDF analysis, and document chat used to be separate things with separate ways to start. That had a cost: comparing a lecture against a textbook chapter on the same topic meant analyzing both separately, opening two different results, and cross-checking them by eye. The notebook removes exactly that step. The same field takes a YouTube link or an article link, the service tells them apart on its own, and a PDF uploads the same way. The difference between "analyze a video" and "analyze a webpage," in terms of what you physically do, is gone: it's the same gesture, paste a link.

How it works technically: from a link to a shared chat

Each source goes through its own breakdown independently of the others: text extraction (captions or speech recognition for video, the text layer for a PDF, the main readable text for a web page), then structuring into a TL;DR, summary, and key points. After that, the source joins the notebook's shared search index: the text gets cut into chunks of roughly 500 tokens with a 50-token overlap, and each chunk becomes a vector through the text-embedding-3-small embedding model. When you ask a question in chat, it becomes a vector too, and retrieval searches across every source in the notebook at once, not just one: up to eight of the nearest chunks go to the model, each tagged with which source it came from. The model also sees the last ten messages of the conversation and the notebook's overall context, up to 20,000 characters, so a follow-up question doesn't need to re-explain what the previous one was about.

How document chat works: the mechanics of semantic search

One chat across several sources: where this actually changes things

"One shared chat across several sources" sounds like a technical detail in the abstract. In practice the difference shows up on a specific task. A marketer pulls five competitor posts into one notebook and one question in chat, what does each competitor emphasize and where do they disagree, gets a comparison across all five at once, instead of reading five separate breakdowns and holding the comparison in their head. A student adds a lecture recording and a textbook chapter on the same topic to one notebook and asks where they disagree: it surfaces exactly what the professor emphasized that the book treats as a footnote, and the reverse. An analyst pulls several quarterly reports into one notebook and asks what changed quarter over quarter, instead of flipping through four PDFs hunting for the paragraph that moved. In all three cases the task is the same: not understanding one document, but comparing several, and that's exactly what a notebook does in one question instead of manual cross-checking.

The notebook and the second brain: different layers of memory

It's easy to conflate the notebook with the second brain, since both are about not losing thoughts, but they solve problems at different scales. A notebook is a workspace around a specific task: several sources, one chat, while you're working through them. It doesn't connect its sources to sources in a separate, different notebook, the same way notebooks in NotebookLM sit isolated from each other. The second brain operates at a different level: individual thoughts you explicitly highlighted and saved, connected by meaning regardless of which notebook or analysis they came from. A thought saved from a video last week can turn out to connect to a thought saved today from an article, if they're genuinely about the same thing, even though they formally belong to two different notebooks and two different sources. The notebook is about the task in front of you right now. The second brain is about what builds up over time on top of any number of those tasks.

What Cruxly's second brain is and how it differs from the notebook

Guest access: try it with no sign-up at all

The first source in a notebook can be analyzed with no sign-up whatsoever: no email, no account, no card on file. Guest access gives one source a day and two chat questions about it, enough to tell whether the chat actually answers on the merits and whether the tool is worth further attention before creating anything. This isn't a stripped-down demo mode: guest analysis runs on the same logic as analysis on any plan, the difference is only the daily limit and that the result isn't saved without an account. NotebookLM, by comparison, requires a Google account from the very first analysis, even on its free tier.

NotebookLM, and why it's worth talking about at all

Ignoring NotebookLM in a conversation about AI notebooks would be strange: it's the product that defined the category, and by third-party Google Trends analysis, interest in it already outpaces interest in Gemini itself at Google. A whole genre of "best NotebookLM alternatives" articles has grown up around it, and the question of how Cruxly differs eventually comes up for anyone who's already tried NotebookLM and is looking for something else. The real difference is access and memory: Cruxly works from Russia with no VPN, and the second brain connects thoughts across different notebooks instead of locking them inside one.

A detailed comparison: Cruxly versus NotebookLM

The main difference for Russia: access with no VPN

NotebookLM physically doesn't open from Russia without switching IP through a VPN: the site can load, but its functionality doesn't work. That's not an abstract shortcoming, it's a concrete barrier that's already spawned its own genre of Russian-language content around NotebookLM, VPN guides, proxy services, workaround payment schemes. Cruxly works from Russia directly, no VPN, with a Russian card. For the task of pulling several sources into one notebook and getting a shared chat across them, that removes a barrier NotebookLM doesn't solve at the product level at all, only at the level of working around a block.

NotebookLM without a VPN: what to do if you need an AI notebook from Russia

The second brain versus siloed notebooks

Another difference is about memory between notebooks, not inside one. In NotebookLM, based on public reports, each notebook lives on its own, separate from the rest: what you saved or discussed in one doesn't automatically surface in another, even if the topics overlap. Cruxly's second brain works the opposite way: a thought saved from a notebook about one project can turn out to connect to a thought from a notebook about an entirely different project, if they're genuinely about the same thing by meaning. That's not the same as a shared chat inside one notebook, it's the next level up: connections that outlive a specific task and a specific notebook, instead of resetting every time a new workspace opens.

The Chrome extension: a second way to start

The notebook isn't the only entry point. The Cruxly Chrome extension shows an "Analyze" button right on a YouTube video, a PDF, or an ordinary article while you're already viewing it, with no trip to a separate site. It's a separate, simpler tool: one source per session, with its own language settings (12 options) and detail level, and a Save button that lands the result in a library on the site, exportable to PDF, Markdown, or DOCX on the Pro plan. If a source analyzed in the extension turns out to belong with others on the same topic, it can be added straight into any notebook from the extension itself, no re-analysis needed. The notebook and the extension solve different situations with the same analysis: the notebook for pulling several sources together, the extension for analyzing one source right where you found it, with precise control over language and depth.

What this looks like in practice: three scenarios

A student is prepping for finals across a dozen lectures. Instead of rewatching each one, they add them as sources to one notebook as the course progresses, and a couple of days before the exam, one question in the shared chat, which topics came up across several lectures, is enough to spot which ones are actually worth a full rewatch. A researcher who relocated to Russia off a project built around NotebookLM looks for a way to keep working without a VPN: they move a dozen papers and interview transcripts into a Cruxly notebook and find the shared, multi-source chat works on the same logic they were already used to. A consultant running several client projects in parallel keeps one notebook per project and saves the important thoughts from all of them into the second brain: when an idea comes up on a new project that was already discussed with a different client six months earlier, it surfaces on its own instead of getting lost inside one notebook's boundaries.

Going deeper: how the notebook handles different source types together

Mixing source types in one notebook isn't the same as mixing different tools. Video yields text through captions or speech recognition, a PDF through its text layer, a web page through extracting the main readable text with the same class of technology behind a browser's reading mode. All three come out the other end as the same kind of data: a structured breakdown plus chunks of source text for retrieval. For the shared chat, that means a question doesn't "know" in advance which source the answer will come from: if a notebook holds a lecture recording, a textbook chapter in PDF, and a news article on the same topic, semantic search pulls relevant chunks from all three with no distinction for whether it's a video, a document, or a web page. The practical upshot: there's no need to decide in advance which source to expect an answer from, just ask directly, and the notebook finds where to look.

How to phrase questions in the shared chat to get precise answers

The more specific the question, the more precisely it retrieves. "Tell me what's in the notebook" works poorly: the entire content of every source matches it equally, and the answer comes back as a generic recap instead of anything concrete. It's more useful to ask something pointed: "where do these sources disagree," "pull together every figure mentioned across the reports," "compare what the video and the article say about the same question." If you know which source an answer should come from, name it directly: "what does the article say about this, not the video" retrieves differently than just "what's known about this," because naming the source becomes part of what retrieval matches against. Split compound questions that pack two different asks into one: that keeps retrieval from straddling two topics and finding worse chunks for both.

Common mistakes when working with a notebook

The first: analyzing several sources one at a time instead of adding them to one notebook from the start. The output is the same either way, but the comparison ends up done by hand, giving up the entire reason the notebook exists. The second: looking for a language or detail-level choice inside the notebook itself, instead of opening the Chrome extension right away, where those settings actually live. The third: conflating the notebook with the second brain and being surprised a thought from one notebook doesn't surface in another notebook's chat, when that's expected behavior — connections between notebooks are a separate feature, not something the chat itself does. The fourth: asking the shared chat something a source's own breakdown already answers, instead of checking the breakdowns first.

Data and privacy

Each source in a notebook gets processed to build its breakdown and join the shared search index: the text goes to a model through a provider, since otherwise it couldn't answer from the content. The files themselves (video, PDF) aren't stored on Cruxly's servers, only the extracted text and the breakdown built from it. A notebook with unsaved sources, and its search index, are deleted automatically after 24 hours if you're not signed in; for a signed-in account, sources stay until manually deleted. The sane rule is the same one that applies to any service in this category: don't upload documents under a real NDA or other formal protection anywhere, including here.

Read our Privacy Policy

Pricing

Every plan and what's included lives on its own page.

Cruxly pricing

The notebook doesn't exist to replace analyzing one file, the old separate analyzers already did that. It exists for the job that starts where one document ends: pulling several sources together and getting one answer instead of several you'd otherwise have to cross-check by hand. Build your first notebook out of whatever you're already working on, and the difference shows up on the very first question that touches more than one source.