What this covers

This is a walkthrough by Nate, a YouTube creator, of a knowledge-base method that Andrej Karpathy shared as a prompt. Nate sets it up live with Obsidian, the Obsidian Web Clipper extension, and Claude Code, in roughly five minutes. The pipeline has three stages: ingesting source documents, having the model organize them into markdown files with explicit relationships, and querying the resulting wiki. He then shows two wikis of his own, one built from YouTube transcripts and one acting as a personal second brain.

The middle of the video is practical detail. Ingesting an article can include clarifying questions from Claude Code about emphasis and granularity. The generated pages carry tags, author attributions, and links to people, organizations, and concepts. The default folders are analysis, concepts, entities, and sources. Karpathy meant the structure to stay flat and his prompt to stay vague, and Nate's YouTube project departs from both with subfolders. A hot cache file holds recent context, a Claude.md file documents how the wiki works, and a periodic "lint" pass finds inconsistencies, fills gaps through web search, and suggests new connections. Obsidian's graph view shows where hubs form and which notes sit isolated. Nate also tells his executive-assistant project to read the wiki only when context is missing.

The closing section compares the approach with RAG, using a chart Nate had Claude generate across infrastructure, cost, maintenance, and scalability. His argument is that the wiki finds information by reading indexes and following links rather than by similarity search, and that it needs no embeddings and no re-embedding when content changes. He cites an X user who cut token usage by 95% across 383 files and over 100 meeting transcripts. He concedes that it breaks down at enterprise scale, with millions of documents. A brief, flagged-as-contested aside covers social-media talk about agentic software being built from high-level ideas given to code agents.

Sharpest takeaway

The speaker demonstrates how to build a personal knowledge management system using Andrej Karpathy's LLM Wiki approach—organizing raw documents into a structured markdown-based wiki with automated relationship mapping—which is simpler, more token-efficient, and more practical than traditional RAG for small to medium-scale knowledge bases.

  • Karpathy's method uses simple markdown files with LLM-guided organization rather than requiring complex vector databases or embedding models
  • The approach reduces token usage by 95% compared to unstructured context while maintaining relationship discovery and querying capability
  • For typical knowledge bases (hundreds of pages), this method outperforms traditional RAG in simplicity and cost; traditional RAG becomes preferable only at enterprise scale (millions of documents)

The claims · ranked33 claims · weighted by value

This asset isn't compiled yet

You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.

0.75

The setup for the Karpathy LLM wiki requires only 5 minutes and does not need a fancy vector database, embeddings, or complex infrastructure—it is literally just a folder with markdown files.

factualhigh valueestablishednovelty 2/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“It's also super simple. It will take you 5 minutes to set up. I'll show you guys. You don't need a fancy vector database, embeddings, or complex infrastructure. It's literally just a folder with markdown files. That's it.”

0.75

The Karpathy LLM knowledge base method has three main stages: data ingest (loading source documents like PDFs), organization by LLM into markdown files with relationships, and Q&A querying through the structured wiki.

definitionhigh valueestablishednovelty 2/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“So, there's different stages. The first part is data ingest. He puts in basically source documents. So, he basically takes a PDF and puts it into Cloud Code, and then Cloud Code does the rest.”

0.75

Karpathy recommends running periodic LLM health checks ('linting') on the wiki to find inconsistent data, impute missing information with web searches, and identify interesting connections for new research areas.

factualhigh valueestablishednovelty 2/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“So, another thing that I alluded to but didn't really cover was the idea of linting. So, Karpathy says that he runs some LLM health checks over the wiki to find inconsistent data, impute missing data with web searches, find interesting connections for new article candidates, things like that.”

0.69

The AI-organized wiki pages include automatically generated tags, author attributions, and hyperlinks to related entities (people, organizations, concepts) that were extracted and connected from the source material.

factualhigh valueestablishednovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“So, let's just see. If I click into the main source, we can see the tags that it got, we can see the authors, and we can click around. So, here's a link to OpenAI. Okay, what's OpenAI? Here's references in AI 2027. Here's some other connections with OpenAI like model spec.”

0.69

Normal AI chats are ephemeral and knowledge disappears after the conversation, but Karpathy's LLM wiki method makes knowledge compound like interest in a bank because it is persistent and organized.

causalhigh valueestablishednovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“Because normal AI chats are ephemeral, meaning the knowledge disappears after the conversation. But, this method using Karpathy's LLM wiki makes knowledge compound like interest in a bank.”

0.69

The graph view in Obsidian shows where the most important content hubs emerge (major nodes with many connections) versus isolated individual nodes, revealing the relationship structure of the knowledge base.

definitionhigh valueestablishednovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“So, I'm going to switch over over here to the graph view because I think it'll be interesting to see as it is starting to go through and create those different wiki files, it's going to go ahead and it's going to create all those relationships, and we'll be able to watch it in real time.”

0.69

Karpathy intentionally left his prompt vague so that users can customize and adapt it to their own project contexts and needs.

factualhigh valueestablishednovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“And Karpathy even said, 'Hey, you know, I left this prompt vague so that you guys can customize it.'”

0.69

The Karpathy LLM wiki method has near-zero cost (only token costs for querying) versus traditional RAG which incurs ongoing compute and storage expenses.

factualhigh valueestablishednovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“The cost over here is basically free. Your only cost is going to be tokens. Whereas, over here you might have ongoing compute and storage.”

0.69

Karpathy intentionally designed the approach to be kept 'really flat' with no subfolders and minimal overorganization, though the speaker shows his YouTube project customizes this with organized subfolders based on project context.

factualhigh valueestablishednovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“Because it's interesting in my personal kind of second brain, the wiki is literally just markdown files. There's no structure to it. And in some cases that's good. Karpathy actually said sometimes I like to keep it really simple and really flat, which means like no subfolders and not a bunch of over organizing.”

0.69

The video demonstrates setting up a complete knowledge management system in 5 minutes using Obsidian, the Obsidian Web Clipper extension, Claude Code, and the Karpathy prompt.

factualhigh valueestablishednovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“What you're looking at right here is 36 of my most recent YouTube videos organized into an actual knowledge system that makes sense. And in today's video, I'm going to show you how you can set this up in 5 minutes. It's super, super easy.”

0.69

The Karpathy wiki method requires only running a lint operation for maintenance (cleaning up inconsistencies and adding articles) rather than re-embedding when content changes.

factualhigh valueestablishednovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“And for maintenance, you just run a lint. You clean up things. You add more articles. You give it more context, rather than having to re-embed when things change.”

0.64

One X user organized 383 scattered files and over 100 meeting transcripts into a compact wiki and reduced token usage by 95% when querying with Claude.

normativehigh valueestablishednovelty 1/4durability 2/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“One X user turned 383 scattered files and over 100 meeting transcripts into a compact wiki and dropped token usage by 95% when querying with Claude.”

0.63

People on social media are discussing how 2026 AI agentic software and products will be built by simply giving high-level ideas to AI code agents rather than following traditional development processes.

forecasthigh valuecontestednovelty 2/4durability 2/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“And people on X are now talking about like this is how 2026 AI agentic software and products will be made. You just give it a high-level idea and it goes and builds it out.”

0.63

Obsidian is a free tool that lets users visually see and navigate their markdown files, serving as an IDE for the knowledge base system.

definitionhigh valueestablishednovelty 0/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“He uses Obsidian as the IDE. So, this is nothing really too game-changing. Obsidian just lets you visually see your markdown files.”

0.63

The Obsidian Web Clipper browser extension allows users to capture articles from the web and directly add them to the Obsidian vault without manual copying or reformatting.

definitionhigh valueestablishednovelty 0/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“Or we can just get an Obsidian extension, which lets us basically take articles right from the web and just put it right into our vault super easy. So search for this extension called Obsidian Web Clipper.”

0.52

The LLM wiki method finds information by reading indexes and following links rather than using similarity search, providing deeper understanding of relationships through explicit connections rather than semantic similarity.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Nate (YouTube Creator)

“So, it finds it by reading indexes and follows links, rather than using similarity search. getting a deeper understanding of relationships, because they're links, rather than just saying, 'Hey, these chunks seem similar.'”

0.48

The LLM wiki can be queried directly within the Obsidian environment or the folder can be pointed at by external AI projects to provide context without embedding the data locally.

factualhigh valuespeaker onlynovelty 1/4durability 3/4· Nate (YouTube Creator)

“So, the question now is like, what do we do from here? Do we query it inside of this environment? Do we query it from somewhere else? And that's completely up to the way that you want to use this. So, for example, with my YouTube project, I'm probably just going to keep this here. And whenever I want to ask questions about YouTube, or if I want to turn this into like a website, I can just do that from here. Or, if I need to, I can point a different project at this folder, since everything's here. And it can crawl through the wiki, it can read the index, and it knows how this stuff works because you can give it the claw.md so it understands the project as well.”

0.48

The speaker provides specific instructions in the executive assistant's claw.md telling it not to read from the wiki unless necessary, to check specific domains first, and to only access wiki data when context is missing.

normativehigh valuespeaker onlynovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“And I said, 'Don't read from the wiki unless you actually need it.' So, here are some things that you might do that you don't need to go read the wiki for. And all of this is my business knowledge.”

0.48

The speaker asked Claude Code to create a comparison chart between the Karpathy LLM knowledge wiki method and traditional semantic search RAG, with specific columns for infrastructure, cost, maintenance, and scalability.

factualhigh valuespeaker onlynovelty 1/4durability 3/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“And the answer is no, but kind of yes. And it all depends on the goal of the project and the goal of the context, how much context you have. So, here's a really quick chart that I had my cloud code make. I was in my Herc brain, where I dumped in a bunch of information about Karpathy's LLM knowledge, and I just said, 'Hey, can you please explain Karpathy knowledge as simple as possible? Keep it super concise, and um compare it to typical semantic search rag.'”

0.45

The system creates different folder structures (analysis, concepts, entities, sources) within the wiki by default, which can be modified based on project needs.

factualhigh valuespeaker onlynovelty 1/4durability 3/4· Nate (YouTube Creator)

“Once we start to populate stuff we can talk to it to see if that's actually the way we want to do it or not. Because it's interesting in my personal kind of second brain, the wiki is literally just markdown files. There's no structure to it. And in some cases that's good. Karpathy actually said sometimes I like to keep it really simple and really flat, which means like no subfolders and not a bunch of over organizing. But then you guys did see in my YouTube transcript one there were different subfolders and I think that in this case it actually makes more sense.”

0.45

A hot cache (hot.md) stores the most recent 500 words or characters of information, helping reduce the need to crawl multiple wiki pages in scenarios where recent context is most valuable.

factualhigh valuespeaker onlynovelty 1/4durability 3/4· Nate (YouTube Creator)

“So, the thing about the hot cash, right? I didn't actually have this in my YouTube one. So, if I go to YouTube, you can see there's no hot cash. But if I go to the Herc brain in the wiki, you can see there's a hot.md right here. And this is basically just a cache of like 500 words or 500 characters that it saves, which is like what is the most recent thing that Nate just gave me or that we talked about. In the context of my executive assistant, this is really helpful, you know, it might save me from having to crawl different wiki pages.”

0.45

The primary weakness of the LLM wiki approach is that it does not scale well across enterprises handling millions of documents, where traditional RAG, knowledge graphs, or light RAG systems become more cost-effective.

causalhigh valuespeaker onlynovelty 1/4durability 2/4· Nate (YouTube Creator)

“But right now, the weakness, of course, with the uh LLM knowledge wiki is that it doesn't scale huge across enterprises, right? Because it's just a bunch of files. Um and that is where the cost will probably get more and more expensive than going to something like standard semantic search, or knowledge graph, or light rag, or whatever other tool is out there for that. So, here you can see, if you have hundreds of pages with good indexes, you're fine with wiki graph. But if you were getting up to the millions of documents, then you're going to want to actually do more of a traditional rag pipeline.”

0.42

Claude Code can ask clarifying questions about article emphasis, focus, granularity, and project plan during ingestion to improve organization and relationship detection.

factualhigh valuespeaker onlynovelty 1/4durability 2/4· Nate (YouTube Creator)

“Okay, so here it came back with all these questions, right? It said, 'Here are my key takeaways from this article, blah blah blah.' And now it'll ask you, 'What do you want to emphasize from this article? What's your focus? How granular do you want to be? What's your plan?'”

0.41

A Claude.md file documents how the project works, how to search through things, and how to update the wiki, serving as comprehensive documentation for the system.

factualhigh valuespeaker onlynovelty 0/4durability 3/4· Nate (YouTube Creator)

“And then, of course, you need your claw.md to explain how the project works and how to search through things and how to, you know, update things.”

0.40

The speaker's personal Herc Brain vault is designed specifically for personal and business information (meeting recordings, ClickUp summaries, Q2 initiatives, employee information) separate from his YouTube research vault.

factualhigh valuespeaker onlynovelty 0/4durability 2/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“And the crazy part is, I said, 'Hey, Claude Code, go grab the transcripts from my recent videos and organize everything.' I literally didn't have to do any manual relationship building here. And then right here, I have a much smaller one, but this is more of my personal brain. So, this is stuff going on in my personal life. This is stuff going on with, you know, Up-to-AI or my YouTube channel or my different businesses and my employees and our Q2 initiatives and things like that.”

0.40

The speaker runs the 36 YouTube transcripts through the Karpathy method in a single batch, which took approximately 14 minutes and created 23 wiki pages.

normativehigh valuespeaker onlynovelty 0/4durability 2/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“When I uploaded the 36 YouTube transcripts in batch, it took about 14 minutes, so it kind of just depends. But, it created 23 wiki pages. We have the source, we have six people, five organizations, and one AI systems page.”

0.39

The system can be monitored in real-time through Obsidian's graph view as Claude Code creates wiki pages and relationships during ingestion.

factualhigh valuespeaker onlynovelty 0/4durability 3/4· Nate (YouTube Creator)

“So, I'm going to switch over over here to the graph view because I think it'll be interesting to see as it is starting to go through and create those different wiki files, it's going to go ahead and it's going to create all those relationships, and we'll be able to watch it in real time.”

0.39

A log file tracks the operation history of the wiki, recording every ingestion and update to the system.

factualhigh valuespeaker onlynovelty 0/4durability 3/4· Nate (YouTube Creator)

“And then we also have a log, which is the operation history. So, in this case, in the YouTube project, the log isn't huge, cuz I only ran one huge batch of the initial 36 YouTube videos. But now, every time I have one, I say, 'Hey, can you go ahead and ingest the new YouTube video into the wiki?' And then we'll see every single time we update this.”

0.38

The ingestion of the AI 2027 article created approximately 25 wiki pages from a single source due to the richness and complexity of the content.

normativehigh valuespeaker onlynovelty 0/4durability 2/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“Okay, so it's creating all of the wiki pages now, and you can see that it said it's going to make about 25 because there's so much stuff going on in the original AI 2027 article.”

0.34

People on X are calling the LLM wiki approach a game-changer because it makes AI feel like a tireless colleague who actually remembers everything and keeps it organized.

factualhigh valuespeaker onlynovelty 0/4durability 1/4· Nate (YouTube Creator)

“People on X are calling it a game-changer because it finally makes AI feel like a tireless colleague who actually remembers everything and it stays organized.”

0.32

Karpathy mentioned that at his scale of about 100 articles and half a million words, he did not need to use fancy RAG infrastructure and the LLM could effectively handle auto-maintaining index files.

normativehigh valuespeaker onlynovelty 0/4durability 1/4· Nate (YouTube Creator)

“So, right now he's doing about 100 articles and about half a million words.”

0.24

The speaker describes an AI 2027 article as an example source, then demonstrates ingesting it by using the Web Clipper to add it to the raw folder and asking Claude Code to organize it into the wiki.

factualestablishednovelty 0/4durability 2/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“Okay, so I'm at this website called AI 2027. If you guys haven't read this before, it's kind of an interesting read, so go check it out. And now let's say I want to get this into my vault. What I could do is just copy the whole page, right? And it might just come through a little weird. Or we can just get an Obsidian extension, which lets us basically take articles right from the web and just put it right into our vault super easy.”

0.19

Andrej Karpathy recently released a post about using LLMs to build personal knowledge bases for research topics, which gained significant traction on social media.

factualestablishednovelty 0/4durability 1/4· Unidentified Speaker — Andrej Karpathy Just 10x’d Everyone’s Claude Code [sboNwYmH3AY]

“So, Andrej Karpathy just released this little post about LLM knowledge bases and explaining what he's been doing with them. And in just a matter of a few days, it got a ton of traction on X.”