LLM Wiki — Personal Knowledge Base
Build a persistent, LLM-maintained wiki that compiles, cross-references, and synthesizes knowledge from your raw sources.
RAG rediscovers knowledge from scratch every time you ask. What if your LLM actually built and maintained a real wiki — compiling sources once, cross-referencing everything, and getting smarter with every document you add?
Who it's for: researchers, PKM enthusiasts, Obsidian power users, knowledge workers, writers with large reference libraries, anyone building a personal knowledge base
Example
"Ingest this 40-page AI safety paper" → Summary page created, 3 entity pages updated, 2 new concept pages generated, 12 cross-references added to existing wiki, 1 contradiction flagged with a prior source
CLAUDE.md Template
New here? 3-minute setup guide → | Already set up? Copy the template below.
npx ccplaybooks add llm-wiki# LLM Wiki — Personal Knowledge Base Builder
You are a knowledge-base curator. Your job is to incrementally build and maintain a persistent wiki — a structured, interlinked collection of markdown files — from raw sources the user provides. You don't just retrieve information at query time; you compile, integrate, and keep it current.
## Architecture
The system has three layers:
### 1. Raw Sources (`sources/`)
Immutable, user-curated documents — articles, papers, images, data files. **You read but never modify these.**
### 2. The Wiki (`wiki/`)Get new playbooks like this one
One email a week with new Claude Code workflows. Free, like everything here.
No spam. Unsubscribe anytime.
What This Does
Turns your LLM into a wiki curator. Instead of using RAG to re-derive answers from raw documents every time, this playbook has Claude incrementally build a persistent wiki — a structured collection of interlinked markdown files that sits between you and your raw sources. Every time you add a new source, Claude reads it, creates summary pages, updates entity and concept pages, adds cross-references, and flags contradictions. Knowledge compounds over time rather than being rediscovered from scratch.
Inspired by Andrej Karpathy's LLM Wiki pattern.
Quick Start
Step 1: Create Your Project Folder
my-wiki/
├── CLAUDE.md ← download the template here
├── sources/ ← drop your raw documents here
│ ├── articles/
│ ├── papers/
│ └── data/
└── wiki/ ← Claude builds this automatically
├── index.md
└── log.md
Create the sources/ and wiki/ directories. Add empty wiki/index.md and wiki/log.md files.
Step 2: Download the Template
Click Download above, then move the file into your project folder as CLAUDE.md.
Step 3: Add Your First Sources
Drop articles, papers, notes, or data files into sources/. Then open Claude Code and say:
Ingest all files in sources/
Claude will process each source — creating summaries, entity pages, concept pages, and cross-references.
Step 4: Start Querying
"What do I know about [topic]?"
"Compare [X] and [Y]"
"What are the gaps in my knowledge about [area]?"
Good answers get filed back into the wiki as exploration pages, so your queries compound over time.
How It Works
The system has three layers:
- Raw sources — immutable documents you curate (articles, papers, data). Claude reads but never modifies these.
- The wiki — markdown files Claude owns entirely: summaries, entity pages, concept pages, comparisons, synthesis. Claude creates, updates, and cross-references these.
- Special files —
index.md(content catalog) andlog.md(chronological operation log) keep everything navigable.
Three Core Operations
| Operation | What it does |
|---|---|
| Ingest | Process a new source → create summary, update entities/concepts, add cross-references, flag contradictions |
| Query | Ask questions → synthesize answers from wiki pages with citations, file good answers back as new pages |
| Lint | Health-check → find contradictions, orphan pages, missing cross-references, knowledge gaps |
Feed the Wiki From Your Reading (Optional)
The wiki grows with every source you add, so the easier it is to capture what you read, the more it compounds.
Option A: Readwise Reader saves articles, PDFs and newsletters in one place and lets you highlight as you read. Export the highlights and drop them in sources/articles/, so Claude builds pages from the parts you marked as important. Readwise does the same for book and Kindle highlights.
Option B: free. Save articles as markdown files or PDFs into sources/articles/ by hand, as in Step 3 above.
The Readwise links are affiliate links: we may earn a commission if you subscribe, at no extra cost to you.
Tips & Best Practices
- Obsidian Web Clipper converts web articles to markdown — fastest way to get sources into your collection.
- Obsidian graph view visualizes your wiki — spot hubs, clusters, and orphan pages at a glance.
- Use git for version history. The wiki is just a folder of markdown files, so branching and collaboration come free.
- Lint periodically. Ask Claude to health-check the wiki every few weeks to catch stale claims and missing connections.
- File answers back. When Claude gives you a great synthesis, say "file this into the wiki" so it compounds.
- For large wikis, consider qmd for local hybrid BM25/vector search with LLM re-ranking.
Why This Works
The tedious part of maintaining a knowledge base is bookkeeping — updating cross-references, keeping summaries current, flagging contradictions, maintaining consistency. Humans abandon wikis because maintenance burden grows faster than value. LLMs don't get bored, don't forget cross-references, and can touch fifteen files in one pass. The wiki stays maintained because maintenance cost drops to near zero.
You curate sources, direct analysis, and ask good questions. Claude handles everything else.