How to Build a Self-Growing Knowledge Base: LLM Wiki for Your Second Brain (Obsidian + Claude) ​

Table of Contents ​

• Most People Using AI for Knowledge Management Are Just Processing Material Temporarily
• The Core of LLM Wiki: Let Material Automatically Grow Relationships
• The Three-Layer Structure of LLM Wiki
• How to Quickly Build a Self-Growing Knowledge Base with Obsidian + Claude Code
• Appendix: Complete CLAUDE.md Rule File and Templates

Most People Using AI for Knowledge Management Are Just Processing Material Temporarily

Most people using AI for knowledge management are essentially repeating the same thing: upload an article → have the AI summarize it → get an answer → close the window.

The next day they upload new material and the AI has to start reading from scratch. No accumulation, no connections, no long-term memory — let alone knowledge compounding.

That's not a real second brain; it's just a "temporary assistant."

Most AI knowledge-base tools on the market today have the same problem: they're more like retrieval systems than truly self-growing knowledge bases. Material gets stored, but it's not deeply digested, connected, or continuously evolving.

The truly advanced approach is: let AI be your knowledge-base administrator. You supply raw material and set direction; AI handles organizing, archiving, cross-referencing, spotting contradictions, and continuous updates. This method is called LLM Wiki, and it delivers a genuinely self-growing knowledge base.

The Core of LLM Wiki: Let Material Automatically Grow Relationships

The biggest difference between LLM Wiki and a typical vector-retrieval knowledge base:

Ordinary tools just chunk and vectorize material, then search and stitch it together when you ask.
LLM Wiki, on the other hand, actively reads each new piece of material and merges it into the existing knowledge network.

It automatically:

• Creates/updates Source pages
• Updates Entity pages (people, companies, products, etc.)
• Updates Concept pages
• Adds cross-references and backlinks
• Flags contradictory viewpoints
• Updates the master index and processing log

For example, if you drop in an article about AI Agent, the AI won't just summarize it — it will:

• Archive the original source
• Extract and update relevant people, tools, and company entities
• Compare, supplement, or correct existing concepts
• Create backlinks
• Log this update

A new piece of material is no longer an isolated note — the whole self-growing knowledge base evolves. That's what a second brain should be.

The Three-Layer Structure of LLM Wiki

Layer 1: Raw Sources
Stores all unprocessed raw material (articles, PDFs, web pages, notes, meeting minutes, etc.). This layer stays read-only to preserve original evidence.

Layer 2: Wiki (the AI-maintained structured knowledge layer)
The knowledge network the AI generates and maintains automatically from raw material, including:

• wiki/sources/ source notes
• wiki/entities/ entity notes
• wiki/concepts/ concept notes

All pages are linked through Obsidian backlinks [[wikilink]], so over time the Graph View grows into an increasingly dense knowledge network.

Layer 3: Schema (the rules layer)
Defined by the CLAUDE.md file: AI's boundaries, workflows, templates, and prohibitions, keeping the knowledge base stable and maintainable long-term.

Obsidian is the IDE, the LLM is the programmer, and the Wiki is the codebase.

How to Quickly Build a Self-Growing Knowledge Base with Obsidian + Claude Code

1. Open Obsidian and create a new vault called test (a local folder).

2. In the Obsidian community plugin store, install Local REST API with MCP, enable the HTTP server, and note down the api key and api url.

3. Open that test folder with Claude Code (or Claude Desktop).

4. Configure the MCP connection (so Claude can operate Obsidian directly).

5. Have Claude Code create the complete structure in one shot:

   请帮我在这个 Obsidian test 里搭建一个 LLM Wiki 自动生长的知识库系统，包括 raw、wiki/sources、wiki/entities、wiki/concepts、templates 等目录，以及 CLAUDE.md 规则文件。

   1
   

6. Put new material into the raw/ folder (naming suggestion: YYYY-MM-DD-short-description.md).

7. Tell Claude Code:

   请读取 raw 里的新文件，并按 LLM Wiki 流程整理进知识库。

   1
   

Claude will automatically summarize, extract entities, distill concepts, update the index, and log changes. Open Obsidian and you'll see the knowledge network growing on its own.

(Refer to the images above for the plugin installation and configuration process)

Appendix: Complete CLAUDE.md Rule File and Templates

CLAUDE.md (Core Rule File) ​

# CLAUDE.md — LLM Wiki 自动生长的知识库规则

## 核心原则
你是一个 LLM Wiki 知识库的专职维护者。你的工作是将**原始资料**转化为**结构化、可增长的知识网络**。

### 铁律：一次只做一件事
- 严格按照 Pipeline 执行，不跳步、不并行。
- 处理 raw 时只读 raw，整理 wiki 时只操作 wiki。

(The full rules, directory structure, workflow Pipeline, naming conventions, link conventions, prohibited items, etc. follow the original high-quality standard — you can copy and use them directly)

Recommended Template Files (source / entity / concept) ​

At the end of the article you'll find complete source-template.md, entity-template.md, and concept-template.md to keep every note consistent in style for long-term maintenance and search.

Summary: with LLM Wiki, you're no longer just piling material into AI — you have a truly self-growing knowledge base and second brain. Obsidian provides local storage and visual networks; Claude provides intelligent maintenance. Together, every piece of input adds bricks to your knowledge system.

If you want to build your own self-growing knowledge base, create a new Obsidian Vault now and follow this article. You'll find knowledge management entering a brand-new phase.

Keywords: self-growing knowledge base, LLM Wiki, second brain, Obsidian knowledge base, Claude Code knowledge management, AI long-term memory system