LLM Wiki Pattern

What It Is

The LLM Wiki pattern (articulated by Andrej Karpathy) is an approach to personal knowledge management where an LLM incrementally builds and maintains a persistent wiki rather than doing one-shot retrieval from raw documents. The wiki is a compiled, compounding artefact: the LLM synthesises sources into it, cross-references accumulate, and contradictions get flagged. Every new source and every good question makes it richer.

This is distinct from RAG: RAG re-derives knowledge from scratch on every query. The wiki compiles knowledge once and keeps it current.

The Three-Layer Architecture

Layer	Who owns it	What it is
Raw sources	You	Immutable documents (articles, papers, clippings). LLM reads but never modifies. Source of truth.
The wiki	The LLM	Synthesised markdown pages: summaries, entity pages, concept pages, comparisons. LLM creates, updates, cross-references. You read it.
The schema	You + LLM	CLAUDE.md / AGENTS.md — tells the LLM the conventions, structure, and workflows. Co-evolved over time.

Core Operations

Ingest — Drop a source into raw/, ask LLM to process it. LLM reads, synthesises, writes/updates wiki pages, updates index, appends to log. A single rich source may touch 10–15 wiki pages.

Query — Ask questions; LLM reads index → relevant pages → synthesises answer with citations. Good answers can be filed back as new wiki pages — explorations compound just like ingested sources.

Lint — Periodic health-check: contradictions, stale claims, orphan pages, missing cross-references, concepts lacking their own page.

index.md — Content-oriented catalogue. Every page listed with a link and one-line summary, organised by category. LLM reads this first on every query to find relevant pages. Works well up to ~hundreds of pages without embedding-based RAG.

log.md — Append-only chronological record of ingests, queries, lint passes. Each entry starts with a consistent prefix (e.g. ## [YYYY-MM-DD] ingest | Source) so it’s parseable with grep.

Why It Works

The bottleneck in knowledge management is not reading or thinking — it is bookkeeping: updating cross-references, keeping summaries current, noting contradictions, maintaining consistency. Humans abandon wikis because maintenance cost grows faster than value. LLMs don’t get bored, don’t forget to update a cross-reference, and can touch 15 files in one pass. The wiki stays maintained because maintenance cost is near zero.

Human role: curate sources, direct analysis, ask good questions.
LLM role: summarizing, cross-referencing, filing, bookkeeping.

Workflow Tips (from Karpathy)

Stay involved during ingest — read summaries, check updates, guide emphasis. Batch-ingest is possible but less valuable.
File good query answers back into the wiki — they compound.
Use Obsidian Web Clipper to get web articles into raw/ quickly.
Download images locally (Obsidian: Settings → Files and links → Attachment folder = raw/assets/).
Obsidian graph view shows wiki shape — hubs, orphans, cluster density.
Marp (Obsidian plugin) generates slide decks from wiki content.
Dataview (Obsidian plugin) queries YAML frontmatter for dynamic tables.
The wiki is just a git repo — version history and collaboration come for free.
At scale, consider a local search engine (e.g. qmd) with BM25/vector hybrid search for the index.

Relationship to the Memex

Conceptually close to Vannevar Bush’s Memex (1945) — a private, curated knowledge store with associative trails. The LLM solves the part Bush couldn’t: who does the maintenance.