--- title: LLM Wiki Pattern created: 2026-05-25
The LLM Wiki Pattern is a novel methodology for building and maintaining personal or team pkm|knowledge bases using LLMs. Developed as a compounding, persistent alternative to standard rag, it shifts the effort from "retrieving and synthesizing on every query" to "compiling knowledge once and keeping it current."
In traditional RAG or doc-chat interfaces (e.g., NotebookLM, ChatGPT uploads), the LLM starts from scratch for every question. It must find relevant fragments, resolve contradictions, and build a synthesis in real time, forgetting everything once the session ends.
The LLM Wiki Pattern introduces a persistent, interlinked wiki (using markdown files) that sits between the raw sources and the user. The LLM acts as the curator, librarian, and programmer of this wiki:
1. Raw Sources (Layer 1): Immutable files (articles, transcripts, PDFs, images) that serve as the ground truth. The LLM reads these but never edits them.
2. The Wiki (Layer 2): Directory of LLM-generated markdown files, categorised into entities/, concepts/, comparisons/, and queries/.
3. The Schema (Layer 3): Standard rules, file-naming conventions, and metadata structures (e.g., SCHEMA.md) that make the LLM a disciplined wiki maintainer.
index.md: Content-oriented, alphabetical list of all pages and one-line summaries.log.md: Chronological record of all actions (e.g., ## [2026-05-25] ingest | Article Title).