Can AI Automatically Organize My Notes into a Wiki? Karpathy's 'LLM Wiki' Fully Explained

An AI That Actually Remembers — Things Are About to Change

Anyone who's spent time with ChatGPT or Claude has probably felt this frustration: you spend an hour organizing information in one conversation, only to start fresh the next day with an AI that remembers nothing. It's like meeting a stranger every single time. The traditional 'RAG (Retrieval-Augmented Generation)' approach has a similar limitation — it searches your documents from scratch with every query. Knowledge never accumulates; you're just running the same treadmill over and over.

In April 2026, Andrej Karpathy — former Head of AI at Tesla and co-founder of OpenAI — published a solution to this problem. He called it the 'LLM Wiki' pattern. Posted as a GitHub Gist, the idea racked up thousands of stars and forks, igniting conversations across the global AI community.

AI as Your Personal Wiki Writer and Curator

The heart of LLM Wiki is compounding knowledge. Karpathy drew an analogy to software compilation: a programmer writes source code, a compiler transforms it once, and the result runs indefinitely. You don't re-read the source code every time — you use the compiled output. LLM Wiki works on exactly the same principle.

The system has three main components:

• Raw materials (raw/): Your original documents, PDFs, links — anything you want the AI to read
• Wiki (wiki/): Markdown documents that the AI writes and maintains itself
• Schema (rules file): Guidelines that define the wiki's structure and how it should be updated

Whenever you add a new document, the AI reads it, updates the relevant wiki pages, and automatically links related concepts together. Karpathy put it this way: "Obsidian is the IDE, the LLM is the programmer, and the wiki is the codebase." His own wiki has grown to roughly 100 documents and 400,000 words.

How the World Is Reacting

New Zealand: As of 2026, 84% of New Zealand workers already use generative AI in some capacity at work. Yet only 17% of organizations have fully integrated AI into their day-to-day operations. A big part of the gap comes down to the lack of structured knowledge management — and that's exactly where LLM Wiki is drawing attention as a practical, actionable solution.

Korea: The concept is spreading rapidly through Korean developer communities like PyTorchKR and Clien. Startup COO Park Se-hee applied the pattern to 4,825 personal notes accumulated over 22 years. The AI's analysis revealed that 94% of her files were isolated — completely disconnected from everything else. Once LLM Wiki started building links between them, her knowledge finally began to compound in a meaningful way.

Key Takeaways

• LLM Wiki is a personal knowledge management method where AI directly writes and maintains a markdown-based wiki
• Unlike traditional RAG, processed knowledge accumulates and connects over time rather than being discarded after each session
• It operates on a three-layer structure: raw source material → AI-generated wiki → defining rules
• While 84% of New Zealand workers use AI, only 17% have systematic knowledge management in place — making LLM Wiki a compelling practical alternative
• Real-world adoption cases are already emerging in the Korean tech community

Wrapping Up

The defining question of AI use in 2026 isn't "How do we make AI smarter?" — it's "How do we better organize the information AI can access?" Karpathy's LLM Wiki offers a practical starting point that anyone can try, no coding skills required. Why not pick one important document tonight, hand it to your AI, and create the very first page of your own wiki?