Apr 6
Loading video player...
Andrej Karpathy posted about using LLMs to build personal knowledge bases - raw articles go in, an LLM compiles them into an interconnected wiki, and health checks keep everything consistent. It went massively viral. But here's the thing: the most valuable raw data isn't external articles. It's your own conversations with your agents. Every time you work with Claude Code, you make decisions, discover gotchas, learn patterns, and build up context that vanishes when the session ends or the context window compacts. What if instead of losing all of that, every conversation automatically compiled into a structured knowledge base that gets smarter over time? That's what I built and in this video I'll show you how to use it! ~~~~~~~~~~~~~~~~~~~~~~~~~~ - Try InsForge - the open source platform that gives your coding agent a database, auth, storage, AI model routing, and hosting all in one. Free to get started, use code InsForgePromo for a free month of Pro: https://insforge.dev/ ~~~~~~~~~~~~~~~~~~~~~~~~~~ - If you're interested in building your own AI second brain to save yourself hours every week, check out the Dynamous community and the new 4 hour second brain bootcamp: https://dynamous.ai/second-brain-bootcamp - Karpathy inspired Claude Code memory system repo: https://github.com/coleam00/claude-memory-compiler - Karpathy's original Tweet: https://x.com/karpathy/thread/2039805659525644595 - Gist to build your own Karpathy LLM knowledge base: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f ~~~~~~~~~~~~~~~~~~~~~~~~~~ 0:00 Karpathy LLM Knowledge Bases 2:32 How it Works 6:01 Data Flow and Architecture 8:40 My Claude Code Memory System 9:19 InsForge 10:43 Setting Up the System 11:04 Obsidian Setup and Vault Configuration 12:16 Claude Code Hooks 16:51 The Compounding Knowledge Loop 18:35 Outro ~~~~~~~~~~~~~~~~~~~~~~~~~~ Join me as I push the limits of what is possible with AI. I'll be uploading videos weekly - at least every Wednesday at 7:00 PM CDT!