Skip the vector DB: a book that makes LLMs maintain your notes

What it does This repository houses a free, bilingual book—part of the Chinese “Orange Book” series—that treats an Obsidian vault as a living knowledge base maintained by Claude Code. It is a methodology guide, not a software package, distributed as Chinese and English PDFs. The text walks through vault architecture, seven concrete workflows with prompts, and plugin curation to turn Markdown notes into an AI-managed system.

The interesting bit The book’s central argument is that LLMs should act as compilers, not retrievers: instead of building a RAG pipeline over your notes, you let Claude Code directly read, edit, and maintain the Markdown files. The author claims this approach has held up at a scale of roughly 400,000 Chinese characters using only a root CLAUDE.md and per-folder index.md files—no complex MCP configuration required.

Key highlights

• Positions Markdown as the “native interface” for AI agents, citing Manus, OpenClaw, and Claude Code as evidence.
• Borrows Andrej Karpathy’s “LLM as compiler” framing to argue against vector-database retrieval for personal knowledge bases.
• Includes seven copy-paste workflows with specific prompts for Claude Code inside Obsidian.
• Covers practical maintenance concerns: Git versioning, local AI options, custom Skills, and narrowing Obsidian’s 1000+ plugin ecosystem down to four essentials.

Caveats

• This is a book repository, not a tool or library; expect PDFs and methodology, not installable code.
• The README explicitly warns that GitHub’s online preview may fail to fully render the PDF files.
• The primary language is Chinese; the English PDF is available, but the surrounding documentation and series context are not.

Verdict Grab it if you are an Obsidian user ready to let an agent take the wheel of your note archive. Ignore it if you were hoping for a deployable app or a conventional RAG codebase.