Karpathy式AI WikiをObsidianでコード不要で構築する方法

Andrej Karpathy’s viral LLM Wiki idea turns Obsidian into a self-updating second brain. This guide shows the 3-layer system: raw sources, AI-built wiki, and queryable outputs, so non-coders can build a living knowledge base that compounds over time. This is the viral X Post I am talking about: Andrej Karpathy @karpathy · Apr 3 LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating Show more 2.7K 8.5K 55K 19M For me, Andrej Karpathy just made Obsidian feel much bigger than a notes app :) This is coming for someone who love taking notes, building lists. My Notes app itself is full of different lists. His LLM Wiki idea gives regular people a simple way to turn raw files into a living knowledge base that writes itself, updates itself, and becomes easier to query over time. The reason this landed so hard is that it removes the biggest barrier most people face: coding. You do not need to build a complex RAG system, wire up scripts, or become an automation nerd to start. You just need a raw folder, an AI-generated wiki layer, and a place to ask questions from inside the system. Why Karpathy’s idea matters Karpathy’s framing is the part that makes the idea click: Obsidian becomes the interface, the LLM becomes the worker, and the wiki becomes the evolving output. So if you understand, instead of dumping your notes into a static archive, you create a system where the notes are continuously reorganized into useful pages, summaries, and answers. That changes Obsidian from a storage app into a real second brain. This is why the trend is catching on so quickly in Obsidian circles. It is not just another productivity trick. It is a better mental model for how knowledge should behave: ingest raw material, compress it, connect it, and make it queryable later. The three-layer system So from reading all the articles, Obsidian etc, what I understood is, the whole workflow is built around three layers: Raw folder This is where you dump source material: PDFs, transcripts, articles, meeting notes, screenshots, and anything else you want the system to learn from. AI-built wiki The AI reads the raw material and turns it into structured wiki pages, topic pages, summaries, and linked notes. This is where the knowledge becomes organized instead of just stored. Queryable outputs Once the wiki exists, you can ask questions, pull summaries, and surface answers from the knowledge base instead of hunting through files manually That layered design is what makes the system feel self-updating. New material enters the raw folder, and the wiki layer evolves on top of it. You are not rebuilding your brain every time you add a note. How to set it up in Obsidian You do not need code for the basic version. Start by creating a vault structure that separates raw material from polished knowledge. Step 1: Create three folders Create these folders in Obsidian: 00 - Inbox 01 - Raw Sources 02 - Wiki Use 00 Inbox for quick capture, 01 Raw Sources for things the AI should process, and 02 Wiki for the cleaned, organized pages. This keeps the workflow simple and prevents your vault from becoming a messy dumping ground. Step 2: Dump everything into Raw Sources Whenever you read an article, watch a video, or get an important email, save the text or notes into 01 Raw Sources. The more consistently you feed this folder, the more useful the wiki becomes. Examples of good raw sources: Article summaries. YouTube transcripts. Book notes. Meeting notes. Research snippets. Do not worry about formatting here. Raw material is supposed to be ugly. The AI layer is what cleans it up. Step 3: Create wiki pages from the raw material For each cluster of related notes, create a page in 02 Wiki. A page might be a topic, a project, a person, or a concept. The point is to convert scattered information into one coherent entry. For example: LLM Basics My Content System Client A Marketing Ideas Obsidian Workflow Each page should summarize the topic, link related pages, and include the most important facts from the raw sources. That is the heart of the wiki effect. The no-code version of the workflow The no-code setup is basically this: You add a note or source into 01 Raw Sources. You ask an AI to summarize, organize, and extract the important parts. You paste that output into a page in 02 Wiki. You keep linking related pages as the wiki grows. That may sound simple, but it is powerful because it turns a pile of notes into a system. The knowledge starts moving from “captured” to “usable.” A practical prompt for the AI layer would look like this: Read the note in my Raw Sources folder and turn it into a clean wiki page. Summarize the key ideas, define important terms, list related concepts, and suggest links to other pages I should create. If you keep doing that consistently, your wiki will improve every time you add something new. How to make it feel self-updating The key is not automation wizardry. The key is a habit loop. New info goes into Raw Sources. AI compresses it into a wiki page. You update links and headings as needed. The wiki becomes the place you return to, not the raw folder. That is what makes the system feel alive. The more you use it, the more useful it gets. You are building a knowledge base that compounds instead of decays. Who this is for This workflow is especially good for: Students who need a better way to turn class material into a study system. Creators who want a permanent content brain. Founders who keep collecting ideas, research, and meeting notes. Non-coders who want the benefits of an AI knowledge system without learning to program. Obsidian users who want something more intelligent than flat folders. If you have ever thought, “I save too much information and use too little of it,” this is the kind of setup that fixes that problem. Why non-coders should care Most AI knowledge-base tutorials assume you can code or are willing to build a mini product. Karpathy’s idea is attractive because it does not ask you to do that. It gives you a mental model and a workflow, not a software project. That matters because the best productivity systems are the ones people actually maintain. If a workflow is too technical, it dies after a week. A simple Obsidian system with a raw folder, a wiki folder, and AI-assisted page creation is much more likely to survive long enough to become useful. Final setup in one sentence If you want the shortest version possible, it is this: collect raw material in one Obsidian folder, use AI to turn it into clean wiki pages in another folder, and keep linking those pages until your notes become a living second brain. I'm Kanika (@KanikaBK), specializing in AI tools, emerging trends, and niche applications. Follow for in-depth analyses, strategic insights, and professional updates to elevate your AI knowledge.