LLM Wiki

Build and maintain a continuously compounding personal knowledge base. The LLM does all the heavy lifting — summarizing, cross-referencing, archiving, and bookkeeping — while you focus on collecting sources, exploring, and asking good questions.

Core principle: Don't use the LLM as a search engine. Use it as a knowledge compiler. Instead of re-deriving knowledge from raw documents on every query (RAG), the LLM incrementally builds a persistent wiki — structured, interlinked markdown files that keep gaining value as each source is added and each question is asked.

Prerequisites

• An LLM agent with file read/write capability
• A file system to store the markdown files
• Node.js >= 18 (required by the management scripts)
• Optional: Feishu sources need lark-context + lark-cli (see the "Feishu sources" section); Obsidian for browsing/graph view; Git for version control

When to use