Building a Personal LLM Wiki: The Andrej Karpathy Workflow in Recall

Check out Recall: https://www.recall.it/?t=astro Use the coupon "ASTRO25" for 25% off In this video, I’m showing you how to build an Andrej Karpathy-style personal LLM wiki using Recall - an AI-powered knowledge base that connects your saved videos, articles, PDFs, notes, and podcasts directly with AI models like GPT, Claude, and Gemini. I’ll walk through how Recall works, how to organize your knowledge into a searchable second brain, and real-world workflows for research, deep learning, and AI-powered thinking. ❓ Frequently Asked Questions Is Karpathy's LLM Wiki easier to set up than a no-code tool like Recall? No. Karpathy's LLM Wiki takes a couple of hours to set up: install Claude Code, create a markdown folder or Obsidian vault, write a CLAUDE.md schema file, and seed prompts. Recall takes under two minutes: sign up, install the Chrome extension, paste a URL. How much maintenance does Karpathy's LLM Wiki need compared to Recall? Karpathy's LLM Wiki needs ongoing lint passes, schema updates, and folder curation as it grows. Recall has effectively zero maintenance: ingestion, indexing, linking, and retrieval are all handled automatically. Do I need technical skills to use Karpathy's LLM Wiki, or can a no-code tool like Recall replace it? Karpathy's LLM Wiki assumes comfort with the terminal, markdown conventions, and writing agent instructions. Recall requires no technical skill at all. If you can paste a link, you can use it. What content formats does Karpathy's LLM Wiki support compared to Recall? Karpathy's LLM Wiki only handles markdown, so PDFs, articles, and transcripts need to be converted and cleaned up manually before they enter the wiki. Recall ingests URLs, wiki links, PDFs, browser bookmarks, Pocket exports, markdown files, and plain notes directly, with a one-click Chrome extension. Can Karpathy's LLM Wiki handle YouTube videos as easily as Recall? No. With Karpathy's LLM Wiki you have to pull the transcript yourself, clean it, and file it into the wiki. Recall saves YouTube videos in one click, generates an automatic summary, and later cites the exact timestamp where any claim was made. How does querying Karpathy's LLM Wiki compare to asking questions in Recall? Karpathy's LLM Wiki requires opening Claude Code in your wiki folder and prompting the agent, which re-reads files each session and is bounded by context limits. Recall lets you ask natural-language questions (for example, "Compare everything I've saved about dopamine detox vs deep work") across your entire knowledge base and returns a structured, cited answer instantly. Does Karpathy's LLM Wiki visualize knowledge connections like Recall? Only if you use Obsidian as your front-end and rely on its graph view. Recall has a built-in knowledge graph that surfaces related cards and lets you explore topics visually. Is Karpathy's LLM Wiki more private than Recall? Yes, Karpathy's LLM Wiki keeps files local on your machine, but you also carry the risk of the agent reading sensitive files unless you sandbox it. Recall is cloud-hosted, which trades manual control for far less friction. Who is Karpathy's LLM Wiki best for, and who should choose Recall instead? Karpathy's LLM Wiki is best for tinkerers who enjoy the terminal and want maximum control over schema and agent behavior. Recall is best for anyone who wants the LLM Wiki outcome, an AI that understands their knowledge, without becoming the system's full-time maintainer. Join The "aiholiq" Newsletter: https://aiholiq.kit.com/newsletter Boost your coding skills with Scrimba: https://scrimba.com/?via=AstroKJ —————————————————— Instagram:   / astroo.ai   X: https://x.com/astrokjoseph LinkedIn:   / astrokjoseph   Website: https://astrokjoseph.com