Save Them This Way! AI Processes It Automatically! Obsidian + Claude + VS Code — It's Done with 3 Folders

One-line summary

Traditional RAG Systems are too complex and expensive for individuals, but a simple three-folder structure combined with AI automation creates an efficient "second brain" that processes and organizes knowledge automatically.

---

Why this matters

The AI era generates enormous amounts of content, but without proper organization, this information becomes overwhelming and unusable. Searching through unorganized content wastes thousands of tokens per query and decreases productivity. Traditional RAG (Retrieval-Augmented Generation) systems solve this problem but require complex vector databases and embeddings—overkill for personal use. This framework provides individuals with enterprise-level knowledge management capabilities at a fraction of the cost and complexity.

---

Core concepts

The three-folder architecture

The system is built on a spatial metaphor visualized at 00:04: imagine three rooms in a building, each serving a distinct purpose in your knowledge workflow.

The dual-tool approach

At 03:25, the framework introduces a powerful pattern: using both Obsidian and VS Code simultaneously on the same local Markdown files. This isn't redundancy—it's strategic division of labor.

• Obsidian excels at reading, visualizing connections (graph view), and managing your knowledge as a human
• VS Code enables efficient AI agent interaction through Claude Code, allowing programmatic file manipulation

Both applications work with the same files in real-time, creating a seamless workflow between human curation and AI automation.

Claude MD files: AI personality per folder

Introduced at 05:00, the CLAUDE.md file is the breakthrough mechanism that makes this system work. Each folder contains its own CLAUDE.md file—the first file Claude reads when operating in that directory.

Raw folder's Claude MD detailed at 08:28:

• Enforces no modification of original content
• Defines file naming: YYYY-MM-DD_TITLE.md
• Specifies minimal frontmatter schema (title, type, source, captured, tags, status)
• Acts as a preservation layer

Wiki/Docs folder's Claude MD shown at 09:05:

• Focuses on rephrasing and reconstructing knowledge
• Generates reusable, AI-optimized content
• Facilitates search and application
• Transforms raw data into structured knowledge

Master Index Claude MD:

• Provides top-level guidelines
• Acts as central control point for AI agents
• Manages cross-folder operations

---

How the argument flows

The problem: Token waste in chaos

The opening visual at 00:00 depicts a person frantically managing scattered notes and papers—this is the reality of AI-era content generation. When you ask an AI agent to find something in this mess, it must process everything, consuming potentially 10,000 tokens per query. The unorganized state makes your growing content library a liability rather than an asset.

Traditional solution: RAG systems

At 01:00, the video explains RAG systems: you feed files into a vector database with embeddings, and AI queries this structured knowledge. While effective, this approach requires:

• Complex vector database setup
• Embedding generation and maintenance
• Significant infrastructure costs
• Technical expertise

For individuals, this is using a bulldozer to plant a garden.

The proposed alternative: Structured simplicity

At 01:45, the three-folder structure emerges as the elegant solution. Instead of complex databases, use simple folders with clear purposes. Instead of manual vector embeddings, use Markdown formatting and Obsidian linking. Instead of expensive infrastructure, use local files with AI automation.

Evidence through demonstration

Starting at 03:30, the video demonstrates the actual setup process:

1. Structure creation at 05:08: Claude Code generates the folder structure based on natural language instructions
2. Rule definition at 08:04: Claude MD files establish folder-specific behaviors
3. Content processing at 09:54: Test documents are generated in Raw folder
4. Automated organization at 11:16: A single "정리해줘" (organize) command triggers Claude to process, clean, and redistribute content according to the established rules
5. Verification at 11:51: Documents appear in Wiki/Docs with proper frontmatter, structure, and links

The conclusion: Personal RAG at 1/4 the cost

By the end of the demonstration, the system achieves RAG-like functionality—AI can quickly locate and process relevant information—but at 4-10 times lower token cost. The workflow is automated: drop content in Raw, issue a command, and AI handles the rest. The knowledge base grows organically, with Obsidian's graph view shown at 12:00 visualizing the expanding web of connections.

---

Applying it in practice

Step 1: Initial setup

• Install Obsidian and VS Code on your local machine
• Create a root folder (e.g., "RAG" or "Knowledge Base") on your computer
• Within this folder, create three subfolders: raw, docs (or wiki), and a master-index.md file

Step 2: Configure Claude MD files

• In VS Code, open your folder structure with Claude Code extension active
• Create CLAUDE.md in the raw folder with rules: preserve original content, define naming convention (YYYY-MM-DD_TITLE.md), specify minimal frontmatter
• Create CLAUDE.md in the docs folder with rules: rephrase for reusability, add comprehensive frontmatter, create internal links
• Create CLAUDE.md in the root folder with top-level organizational guidelines

Example Raw folder Claude MD structure:

# Raw Folder Rules
- Never modify original content
- File naming: YYYY-MM-DD_TITLE.md
- Frontmatter: title, type, source, captured, tags, status
- Workflow: capture → minimal processing → move to docs when refined

Step 3: Establish the dual-tool workflow

• Open your knowledge base folder in VS Code for AI interaction and coding
• Open the same folder as a vault in Obsidian for reading and visualization
• Configure Obsidian settings for wikilinks and backlinks
• Test that both applications can see and edit the same files simultaneously

Step 4: Content ingestion process

• When you encounter new information (articles, videos, notes), immediately save to raw folder
• Use a consistent naming convention (date-based helps chronological tracking)
• Don't worry about formatting or organization at this stage—just capture

Step 5: AI-powered organization

• Periodically (daily or weekly), open VS Code with Claude Code
• Issue a command like "Organize the raw folder according to our system rules"
• Claude will read the CLAUDE.md instructions, process raw files, and:
• Clean and format content
• Add proper frontmatter
• Create cross-references and links
• Move refined content to docs folder
• Update the master index

Step 6: Knowledge exploration and growth

• Use Obsidian's graph view to visualize knowledge connections
• Follow backlinks to discover related notes
• Customize Obsidian's appearance with CSS as shown at 13:17 to match your aesthetic preferences
• Let the system grow organically—each new note becomes part of your interconnected knowledge web

---

Caveats & limits

Technical prerequisites

This system requires basic comfort with:

• Installing and configuring desktop applications
• Working with file systems and folders
• Writing simple Markdown
• Interacting with AI through text commands

Users completely new to these tools will need initial learning time.

Consistency dependency

The effectiveness of this system relies entirely on adhering to the structure. If you bypass the Raw folder, save files with inconsistent naming, or ignore the folder purposes, the AI cannot maintain organization. The system works because of its constraints, not despite them.

Token costs still exist

While dramatically cheaper than traditional RAG (4-10x reduction), Claude still consumes tokens for processing. Each "organize" command requires Claude to:

• Read all Claude MD files
• Process raw content
• Generate formatted output
• Create links and frontmatter

For very large knowledge bases (thousands of files), periodic organization runs can still incur meaningful costs.

Initial setup investment

Defining effective Claude MD rules requires upfront thought:

• What frontmatter fields matter for your work?
• How should files be named for your retrieval patterns?
• What level of processing belongs in Raw vs. Docs?

This isn't install-and-go—it's build-your-own infrastructure. The payoff comes over time as the system matures.

Not a replacement for true RAG at scale

For enterprise applications with millions of documents, team collaboration requirements, or need for semantic search across massive datasets, traditional RAG systems remain superior. This framework optimizes for the individual knowledge worker, not organizational infrastructure.

---

Quotes to remember

---

Related concepts

• RAG System — Retrieval-Augmented Generation architecture
• Personal Knowledge Management — PKM frameworks and methodologies
• Obsidian — Knowledge base and note-taking application
• VS Code — Microsoft Visual Studio Code editor
• Claude — Anthropic's AI assistant
• Claude Code — VS Code extension for Claude integration
• Markdown — Lightweight markup language for formatted text
• Zettelkasten — Note-taking and knowledge organization method
• Second Brain — Personal knowledge management philosophy
• YAML Frontmatter — Metadata format for Markdown files
• Wikilinks — Internal linking syntax in knowledge bases
• Graph View — Visual representation of note connections
• AI Agents — Autonomous AI systems that perform tasks

---

Key timestamps

• 00:00 — Introduction to the problem of unorganized AI-generated content and the need for a system
• 01:00 — Introduction to the RAG system and its limitations for individuals
• 01:45 — Presentation of the proposed three-folder structure: Raw, Wiki, Master Index
• 02:45 — Explanation of Obsidian and VS Code as tools for managing Markdown files
• 03:30 — Demonstration of setting up the folder structure using VS Code and Claude
• 05:00 — Introduction and demonstration of Claude MD files for folder-specific AI instructions
• 08:00 — Demonstration of adding content to the 'Raw' folder and triggering the AI to process it
• 10:00 — Review of processed documents in Obsidian, showing front matter and structured notes
• 12:00 — Discussion of Obsidian's graph view and further customization options (CSS)
• 13:30 — Concluding remarks and preview of future content