Skip to content
Primores AI wiki

LLM Wiki Pattern — What It Means

LLM Wiki Pattern

TL;DR: Instead of AI retrieving from raw documents each time (RAG), the LLM builds and maintains a structured wiki that compounds knowledge over time. You curate sources and ask questions; the LLM handles all the bookkeeping.

Simple Explanation

Most AI document tools work like this: upload files, ask questions, AI retrieves relevant chunks, generates answer. Repeat. Nothing accumulates.

The LLM Wiki Pattern is different:

  1. You add a source to your collection
  2. The LLM reads it, extracts key information
  3. The LLM integrates it into an existing wiki — updating pages, adding cross-references, noting contradictions
  4. The knowledge is compiled once and kept current
  5. Every source makes the wiki richer

The wiki is the product, not chat history.

Think of it like the difference between:

  • Googling the same thing repeatedly (RAG) vs.
  • Building a personal encyclopedia that gets smarter (Wiki Pattern)

Why It Matters for Business

This pattern solves the “abandoned wiki” problem:

“Humans abandon wikis because the maintenance burden grows faster than the value. LLMs don’t get bored, don’t forget to update a cross-reference, and can touch 15 files in one pass.”

For business applications:

  • Competitive intelligence — Wiki that stays current as you add sources
  • Internal knowledge base — Fed by meetings, Slack, documents; maintained by AI
  • Research projects — Deep dives that accumulate understanding
  • Client knowledge — Everything you learn about a client, synthesized

The Three Layers

LayerWhat It IsWho Owns It
Raw sourcesOriginal documents, articles, dataImmutable — your source of truth
The wikiStructured, interlinked markdown pagesLLM creates and maintains
The schemaRules for how the wiki worksYou and LLM co-evolve

The Three Operations

1. Ingest

Add a source → LLM reads it → Discusses key points → Updates wiki pages → Logs activity

A single source might touch 10-15 wiki pages.

2. Query

Ask a question → LLM searches wiki → Synthesizes answer with citations → Good answers become new wiki pages

Queries compound back into the knowledge base.

3. Lint

Periodically health-check → Find contradictions, orphan pages, gaps → Suggest improvements

Keeps the wiki healthy as it grows.

The Key Insight

“Good answers can be filed back into the wiki as new pages. A comparison you asked for, an analysis, a connection you discovered — these are valuable and shouldn’t disappear into chat history.”

Your explorations compound just like ingested sources do.

Historical Connection

The pattern echoes Vannevar Bush’s Memex concept (1945):

  • A personal, curated knowledge store
  • Associative trails between documents
  • Connections as valuable as documents themselves

Bush’s vision couldn’t solve who does the maintenance. LLMs solve that.

Real-World Example: This Wiki

This very wiki (Primores AI Wiki) implements the LLM Wiki Pattern:

  • raw/ — Source documents (private)
  • wiki/ — LLM-maintained knowledge base (public)
  • methodology — How this wiki is built and maintained

When you’re reading this page, you’re seeing the pattern in action.

Common Misconceptions

  • ❌ Myth: You need to write the wiki yourself

  • ✅ Reality: LLM writes and maintains it; you curate sources and ask questions

  • ❌ Myth: This requires complex infrastructure

  • ✅ Reality: Just markdown files + an LLM agent (like Claude Code)

  • ❌ Myth: This replaces RAG entirely

  • ✅ Reality: Different tools for different jobs; wiki pattern is for compounding knowledge

Tools That Support This

  • tools/obsidian — IDE for viewing/navigating the wiki
  • Claude Code / Codex / similar — LLM agents that maintain the wiki
  • qmd — Local search engine for larger wikis
  • Git — Version history for free

Key Takeaways

  • Wiki pattern = compile knowledge once, keep it current
  • LLM handles all the boring maintenance humans abandon
  • Three layers: raw sources, wiki, schema
  • Three operations: ingest, query, lint
  • Good answers become wiki pages — everything compounds