LLM Wiki Pattern — Helicopter View

The Problem

RAG Is Stateless

Every query starts from zero. No cross-references built. No contradictions detected. No understanding compounds. It is a librarian that fetches books but never reads them.

No Memory No Synthesis

Core Concept

What Is the LLM Wiki Pattern?

Instead of retrieving from raw documents at query time, the LLM incrementally builds and maintains a persistent wiki of markdown files. Knowledge is compiled once and kept current, not re-derived on every query. When you add a new source, the LLM reads it, extracts key information, and integrates it into existing pages. Cross-references are pre-built. Contradictions are flagged during ingest. Every source makes the wiki richer.

New Source> LLM Processes> Updates Wiki> Knowledge Compounds

Mental Model

IDE + Programmer + Codebase

Obsidian is the IDE. The LLM is the programmer. The wiki is the codebase. You browse the results in real time while the LLM makes edits based on conversation.

RAG = search engine for fragments. Wiki = living research notebook.

Layer 1

Raw Sources

Immutable collection of source documents. Articles, papers, transcripts, images. The LLM reads from them but never modifies. Your evidence locker and source of truth.

Immutable Traceable

Layer 2

The Wiki

LLM-generated markdown: summaries, entity pages, concept pages, comparisons, running synthesis. The LLM owns this layer entirely. Knowledge accumulates here.

LLM-Owned Compounding

Layer 3

The Schema

Config doc (CLAUDE.md) defining structure, conventions, and workflows. The meta-layer and control surface. Human and LLM co-evolve this over time.

Co-Evolved Meta-Layer

Architecture

Operation 1

Ingest

Drop a new source. LLM reads it, writes a summary, updates the index, revises 10-15 related wiki pages, appends to the log. One-at-a-time with review is preferred.

Source> Process> Update

Operation 2

Query

Ask questions against the wiki. LLM reads the index, finds relevant pages, synthesizes an answer with citations. Good answers file back into the wiki. Explorations compound too.

Cited Filed Back

Operation 3

Lint

Health-check the wiki. Find contradictions, stale claims, orphan pages, missing concepts, data gaps. Suggests new questions and sources to investigate.

Contradictions Orphans

Comparison

RAG vs LLM Wiki

Dimension	RAG	Wiki
Knowledge	Stateless	Compounds
Cross-refs	Query-time	Pre-built
Setup	Higher	Markdown only
Scale	Millions	Hundreds
Contradictions	Missed	Flagged

Anti-Patterns

What Breaks It

Batch-ingest without reviewing
Neglect the schema document
Skip lint passes entirely
Treat as RAG replacement at scale
Write the wiki yourself

Where It Shines

Best Use Cases for the Pattern

Any domain where knowledge accumulates over weeks or months and cross-referencing matters. The LLM handles the maintenance nobody wants to do. Human thinks, LLM bookkeeps.

Research Deep Dives Book Companion Personal Development Team Knowledge Competitive Intel Course Notes

Get Started

5 Steps to Begin

Pick your LLM agent
Create raw/ + wiki/ + schema
Write a minimal schema doc
Ingest your first source
Build iteratively from there

Just a git repo of markdown files. No database. No vector store.

The LLM Wiki Pattern