How to Structure Your Obsidian Vault So Claude Code Can Actually Find Things

Claude Code can read and write files across your entire Obsidian vault, but only if the vault is organized in a way the model can navigate. This video walks through a specific folder structure called the Carpathic Obsidian RAG method that gives Claude Code a clear map of your knowledge base, then explains why the exact layout matters less than having any consistent system at all.

The Carpathic Obsidian RAG Method

The approach starts with your Obsidian vault as the root directory and breaks everything into three subfolders:

raw/ stores all unstructured data. This is where research notes, bookmarks, article clippings, and rough notes on any topic (like AI agents) go. Think of it as the intake layer where nothing needs to be polished or organized.
wiki/ holds structured, topic-based articles written in a Wikipedia-like style. You take the raw research on a subject and distill it into clean reference pages. The video shows this folder containing a _master-index.md file that acts as the LLM's table of contents, with topic subfolders like ai-agents/, rag-systems/, and content-creation/, each with their own _index.md.
output/ is where finished deliverables land. Reports, slide decks, query results, and any other artifacts that Claude Code produces by drawing on the wiki layer. The video shows files like query-results.md and a slide-decks/ subfolder.

The key detail in the diagram is that Claude Code sits inside the vault. You open it at the vault root, and it works within this file structure. That gives it the ability to locate any document across thousands of files because the hierarchy tells it where to look.

Why "RAG" Is in Air Quotes

Chase calls this a RAG method but flags immediately that it is not a true RAG system. There is no vector database, no embedding pipeline, and no semantic search happening under the hood. Instead, Claude Code relies on the file system itself as the retrieval mechanism. The folder hierarchy and index files replace what a traditional RAG stack would do with embeddings and similarity search. It works because Claude Code can traverse directories and read markdown files directly.

The Real Takeaway: Pick Any System That Works

The second half of the video pushes back on the specific three-folder structure. Chase is direct about it: you do not have to build it exactly like this. The Carpathic method is one good option, but the only requirement is a file structure where Claude Code can find things across thousands of documents. If your system makes sense to you and has enough structure for an LLM to navigate, that is sufficient. The rigid taxonomy of raw, wiki, and output is a starting point, not a requirement.

Key Takeaways

Organize your vault into layers: raw research, structured wiki pages, and finished outputs. This gives Claude Code a clear path from source material to deliverables.
Use index files as navigation aids. The _master-index.md and per-topic _index.md files act as a table of contents that helps Claude Code find the right documents without scanning every file.
This is not true RAG. There are no embeddings or vector search involved. The file system itself is the retrieval layer.
The specific folder names do not matter. What matters is having a consistent, logical hierarchy that both you and Claude Code can navigate at scale.
Claude Code runs from the vault root. Open it at the top level of your Obsidian vault so it has access to the entire file tree.

Resources

Obsidian - the markdown-based knowledge management app used as the vault layer
Claude Code Documentation - Anthropic's official CLI for Claude, which operates on local file systems
Carpathic on GitHub - the creator of the original Obsidian RAG folder structure referenced in the video