llm-wiki

A personal knowledge base continuously maintained with Obsidian + LLM Wiki. Inspired by Karpathy's LLM Wiki pattern.

Core idea: Knowledge is not derived from scratch on every query — it is compiled once and continuously accumulated. Humans curate materials and ask questions; the LLM handles all the tedious organization work — summarization, cross-referencing, archiving, and consistency maintenance.

Core Principles

Compilation over Retrieval

Traditional RAG retrieves from raw documents and assembles answers from scratch on every query — knowledge never accumulates. LLM Wiki takes the opposite approach: each time material is ingested, the LLM compiles knowledge into structured wiki pages, and subsequent queries are answered based on pre-compiled knowledge. Knowledge is compiled, linked, and continuously optimized — just like code.

Knowledge Lifecycle Management

Knowledge doesn't stay valid forever once written. This project introduces a complete lifecycle management mechanism: