LLM Wiki

The existing RAG (Retrieval-Augmented Generation) approach retrieves relevant documents at the time of the question to generate an answer, so knowledge must be rediscovered from scratch every time, and information does not accumulate.

The LLM Wiki is an approach where an LLM reads documents, extracts key information, and manages it as structured Markdown files. Every time a new source is added, the LLM updates the existing wiki, revises entity pages, and identifies contradictions between pieces of information. Through this, knowledge is not re-derived each time but continuously compounds.

The system consists of three layers:

• Raw sources: Original document data that is not modified.
• The wiki: A directory of Markdown files that the LLM creates and manages.
• The schema: A configuration document that defines the wiki's structure and workflows.

Users focus on sourcing and asking questions, while the LLM handles the tedious work such as summarizing, cross-referencing, and record-keeping. When users use Obsidian like an IDE, the LLM manages the wiki as a codebase, like a programmer, making it possible to build a powerful knowledge base across various domains such as personal research, reading, and business knowledge management.