How to Build Your Own LLM Knowledge Base

How to stop starting from zero every time you open AI.

You use AI like a search engine with memory loss. You ask a question, get an answer, close the tab. Tomorrow you ask something related and start from zero again. Nothing builds up. Nothing connects.

Andrej Karpathy talked about using LLMs not as answer machines, but more like librarians. You feed them raw stuff (articles, PDFs, notes, anything), and they quietly build a personal wiki in the background. The wiki grows. Ideas start linking to each other. And when you ask questions later, the AI researches against its own compiled knowledge instead of guessing from scratch.

People call this an LLM knowledge base. Basically your own external brain.

What You’re Actually Building

Two folders. That’s the whole thing.

1. Raw folder — Holds articles you clipped, notes you wrote, PDFs, anything you want to remember later. This folder stays untouched. The AI reads from it but never messes with it.
2. Wiki folder — This is where the AI writes. Summaries of each source, articles about the key concepts, and an index that ties everything together. You basically don’t touch this folder either. It’s the AI’s job now.

You feed the first folder. The AI builds the second folder. You ask questions against the second folder.

Setup (2 Minutes)

1. Download Obsidian (free, works on Mac, Windows, Linux, phone)
2. Open it and create a new vault. Name it something useful (e.g., ai-research, crypto-brain, startup-ideas)
3. Inside the vault, create two folders: raw and wiki

Setup done.

Full article includes: exact prompts for the AI update flow, weekly cleanup routine, useful tools, and the practical system that turns a few saved articles into a knowledge base that actually compounds.