AI Ark LLM Wiki

LLM Wiki: Build a Personal Knowledge Base with Obsidian & Claude Code

閱讀時間約 5 分鐘

LLM Wiki: Build a Personal Knowledge Base with Obsidian & Claude Code

by Teachers Tech (Jamie)

The Problem with RAG

There’s a problem with the way most of us use AI right now. When you upload documents to something like ChatGPT or Notebook LM and ask a question, the AI searches through your files, pulls out some relevant pieces, and gives you an answer. That works. But ask a similar question tomorrow, and the AI does all of that work again from scratch. Nothing was saved. Nothing was built up. Every single question starts from zero.

Andrej Karpathy, co-founder of OpenAI, former AI director at Tesla, recently shared an idea that fixes this problem. He calls it the LLM wiki.

What is LLM Wiki?

Instead of searching raw documents every time you ask a question, you have the AI read your documents once and build a structured wiki out of them — a real persistent knowledge base made of interlinked markdown files.

When you add a new source (PDF or article), the AI:

  • Reads it and extracts key ideas
  • Integrates them into the wiki (updates existing pages, creates new pages for new concepts)
  • Links related ideas together
  • Flags contradictions if new sources conflict with existing wiki content

Over time, the wiki keeps growing and getting richer. The connections are already there. The synthesis is already done. When you ask a question, the AI is not starting from scratch — it’s working from a pre-built organized knowledge base.

Karpathy’s Analogy

“Think of Obsidian as the IDE, the LLM as the programmer, and the wiki as the code base. You rarely write the wiki yourself. The AI does the writing and organizing. You focus on what goes in and what questions to ask.”

Three-Layer Architecture

Layer 1: Raw Sources Your original documents — PDFs, articles, meeting notes. Read-only. The AI reads them but never changes them. This is the source of truth.

Layer 2: The Wiki A folder of markdown files that the AI creates and maintains. Includes: index page, concept pages, entity pages, summary comparisons, all interlinked, all maintained by the AI.

Layer 3: The Schema A rules document (CLAUDE.md) that tells the AI how to structure the wiki, how to handle new sources, how to format everything.

Setup Walkthrough

Tools needed:

  • Obsidian (free, obsidian.md) — markdown viewer
  • Claude Code (or OpenAI Codex, Cursor) — coding agent that can read/write files

Obsidian Setup:

  1. Create a new vault (e.g. “LLM wiki”)
  2. Create three folders: raw/, wiki/, templates/
  3. Create CLAUDE.md in the vault root as the schema file

Schema (CLAUDE.md) contains:

  1. Purpose — what the knowledge base is about (only line to customize per project)
  2. Folder structure — where raw resources and wiki output go
  3. Ingest workflow — read document → extract key concepts → create/update wiki pages → update index → log changes
  4. Page formatting rules — summary at top, claims reference sources, link to related concepts
  5. Question answering behavior — consult wiki first, cite sources, flag uncertainty

Ingest Demo (planning a trip to Japan):

  1. Drop a travel blog article into raw/ (as markdown via Obsidian Web Clipper)
  2. Open Claude Code in the vault directory
  3. Prompt: “I just added a new source to the raw folder. Please read it and update the wiki.”
  4. Claude reads the article, creates structured wiki pages (neighborhood pages, etc.)
  5. Check Obsidian → Graph View shows connections forming
  6. Add second source (food guide) → repeat ingest prompt
  7. Claude updates existing pages (adds food info to neighborhood pages) and creates new ones
  8. Graph View grows denser with more nodes and connections
  9. Ask cross-source question: “What neighborhood should I stay at if I want to be close to the best food and still near the major temples?”
  10. Claude pulls from wiki pages (neighborhoods, food, temples) — not raw articles

Linting:

  • Periodically ask AI: “Please lint the wiki.”
  • Checks for: contradictions, outdated claims, orphan pages (no links pointing to them), concepts mentioned but without their own page
  • Keeps the wiki healthy as it grows

Use Cases

  • Students/Researchers: Build wiki from papers and articles on a topic. End result = structured knowledge base, not a pile of highlighted PDFs.
  • Teachers: Feed in curriculum documents, professional development materials, articles.
  • Business: Feed in meeting notes, customer call transcripts, project documents — new team members can browse organized wiki instead of digging through Slack history.
  • Curious readers: Track what you learn from books, podcasts, and articles — build your own personal encyclopedia.

Limitations

  1. Works best at personal scale (~100 articles). Tens of thousands of pages need more infrastructure.
  2. Garbage in, garbage out — the wiki is only as good as the sources you feed it.
  3. Requires a coding agent (Claude Code, Codex, or similar tool) — Obsidian alone doesn’t do this.
  4. AI can make mistakes (mis-categorize or misconnect) — lint feature exists for a reason.

Summary

The LLM Wiki is a personal knowledge base that the AI builds and maintains for you, that actually gets better over time instead of starting from scratch on every question. Free to set up. Data stays on your computer in plain text files that you own.

Graph View