Llm Wiki

Karpathy LLM Wiki

A pattern for building personal knowledge bases using LLMs.

Based on Andrej Karpathy's LLM Wiki.

The Core Idea

Instead of RAG (retrieve-on-every-query), the LLM incrementally builds and maintains a persistent wiki — a structured, interlinked collection of markdown files. When you add a new source, the LLM reads it, extracts key information, and integrates it into the existing wiki — updating entity pages, revising topic summaries, noting contradictions, and strengthening the evolving synthesis.

The wiki is a persistent, compounding artifact. Cross-references are already there. Contradictions have been flagged. Synthesis reflects everything you've read. It gets richer with every source and every question.

You never write the wiki yourself — the LLM writes and maintains all of it. You're in charge of sourcing, exploration, and asking the right questions.

Architecture (Three Layers)

1. Raw Sources

Your curated collection of source documents — articles, papers, images, data files. These are immutable. The LLM reads from them but never modifies them. This is your source of truth.

2. The Wiki

A directory of LLM-generated markdown files — summaries, entity pages, concept pages, comparisons, overview, synthesis. The LLM owns this layer entirely: creates pages, updates them when new sources arrive, maintains cross-references, keeps everything consistent. You read it; the LLM writes it.

3. The Schema

A configuration document (this skill) that tells the LLM the wiki's structure, conventions, and workflows. You and the LLM co-evolve this over time.

Operations

Ingest

User drops a new source. The LLM:

1. Reads the source and discusses key takeaways
2. Writes a summary page in the wiki
3. Updates the index
4. Updates relevant entity and concept pages across the wiki
5. Appends an entry to the log

A single source may touch 10-15 wiki pages. Prefer ingesting one source at a time for quality.

Query

Ask questions against the wiki. The LLM:

1. Reads index.md to find relevant pages
2. Reads those pages and synthesizes an answer with citations
3. Good answers get filed back into the wiki as new pages — comparisons, analyses, connections should be persisted, not lost in chat history

Lint (Health Check)

Periodically health-check the wiki. Look for:

• Contradictions between pages
• Stale claims superseded by newer sources
• Orphan pages with no inbound links
• Important concepts mentioned but lacking their own page
• Missing cross-references
• Data gaps that could be filled with a web search

Default Location

The wiki lives at a user-level path so it's accessible from any workspace or conversation:

Copy code
~/.workbuddy/wiki-knowledge/

This is the default. The LLM should:

1. Always check ~/.workbuddy/wiki-knowledge/ first when the user mentions their wiki, regardless of the current workspace.
2. If the directory exists and contains WIKI-SCHEMA.md, use it directly — no need to ask the user for a path.
3. Only ask for a custom path if the user explicitly wants a separate wiki.

Getting Started

When a user wants to create a wiki, follow these steps:

Step 1: Initialize Structure

Create at ~/.workbuddy/wiki-knowledge/ (or a user-specified path):

Copy code
wiki-knowledge/
├── raw/              # Immutable source documents
│   └── assets/       # Downloaded images
├── wiki/             # LLM-maintained markdown files
│   ├── index.md      # Content catalog
│   └── log.md        # Chronological operation log
└── WIKI-SCHEMA.md    # Schema & conventions (co-evolve with user)

Step 2: Create index.md

A catalog of everything in the wiki — each page listed with a link, a one-line summary, and metadata (date, source count). Organized by category. The LLM updates it on every ingest.

Step 3: Create log.md

Append-only chronological record. Each entry format:

markdownCopy code
## [YYYY-MM-DD] operation_type | Title
Brief description of what was done.

Use consistent prefixes so it's parseable with grep:

bashCopy code
grep "^## \[" log.md | tail -5

Step 4: Create WIKI-SCHEMA.md

A configuration file describing:

• Directory structure and conventions
• Page format templates
• Cross-referencing rules
• Domain-specific categories
• Workflows for ingest/query/lint

Customize this with the user for their specific domain.

Page Format Template

markdownCopy code
---
title: Page Title
type: entity | concept | source-summary | comparison | synthesis
created: YYYY-MM-DD
updated: YYYY-MM-DD
sources: [list of source filenames]
tags: [relevant tags]
---

# Page Title

Content here. Use [[wiki-links]] for cross-references.

## See Also
- [[related-page-1]]
- [[related-page-2]]

Use Cases

Tips

• Obsidian works great as the wiki viewer — graph view shows connections, Dataview plugin queries frontmatter metadata
• The wiki is just a git repo of markdown files — version history, branching, and collaboration for free
• Marp can generate slide decks from wiki content
• Web Clipper (Obsidian plugin) converts web articles to markdown for easy source ingestion
• Download images locally for persistence — set a fixed attachment directory

Important Principles

1. Raw sources are immutable — never modify source documents
2. The LLM owns the wiki — users read, LLM writes and maintains
3. Knowledge compounds — every ingest and query enriches the wiki
4. File good answers back — don't let valuable analysis disappear into chat
5. Regular lint passes — keep the wiki healthy as it grows