Building a Second Brain with Claude + Obsidian: The Ultimate Guide to Compounding Knowledge

·

4 min read

Building a Second Brain with Claude + Obsidian: The Ultimate Guide to Compounding Knowledge

AI Second Brain Hero

Imagine a world where you never "forget" a great podcast, a crucial insight from a book, or a breakthrough idea from a midnight brainstorming session. Most of us have "digital graveyards"—bookmarks we never reopen and Notion pages that grow stale because the effort to maintain them is too high.

Digital Graveyard

The secret to a truly functional Second Brain isn't a better app or a complex tagging system. It's offloading the bookkeeping to an AI.

By combining Claude Code (the engine), Obsidian (the window), and a Flat-File Schema (the map), you can build a knowledge base that doesn't just store information, but compounds it.


🧠 The Philosophy: RAG vs. Compounding Knowledge

Most people use AI for "Retrieval-Augmented Generation" (RAG). You upload a PDF, ask a question, and the AI finds the answer. But every time you ask, the AI starts from scratch. There is no growth.

The Compounding Approach treats your knowledge base like a living wiki. Instead of just retrieving data, Claude proactively:

  1. Synthesizes new sources into existing pages.
  2. Links related concepts across different files.
  3. Flags contradictions (e.g., "Source A says X, but your new note says Y").
  4. Maintains an index so you always know where things are.

In this system, Obsidian is the IDE, Claude is the programmer, and your Wiki is the codebase.


🛠️ Step-by-Step Setup Instructions

Step 1: The "Simple & Flat" Structure

Forget complex folders. Create one main project folder with three sub-directories. This structure allows Claude to navigate your data without getting lost.

The Knowledge Flow

my-knowledge-base/
├── raw/      # The "Junk Drawer": Articles, screenshots, raw notes, exports. (Immutable)
├── wiki/     # The "Artifact": AI-generated summaries, concept pages, and links.
└── outputs/  # The "Lab": Answers, reports, and synthesized research.

Step 2: Create Your "Instruction Manual" (CLAUDE.md)

The AI needs a set of rules to stay disciplined. Create a file named CLAUDE.md in the root folder. This is the "Schema" that prevents your wiki from becoming a mess.

Copy this starter template into your CLAUDE.md:

# Knowledge Base Schema

## Architecture
- raw/ contains immutable source material. Never modify these files.
- wiki/ contains the organized wiki maintained by the AI.
- outputs/ contains generated reports and analyses.

## Wiki Rules
- Every topic gets its own .md file in wiki/.
- Every wiki file must start with a concise summary.
- Use [[topic-name]] for internal linking.
- Maintain an INDEX.md in wiki/ as a master catalog.
- When new sources are added to raw/, update all relevant wiki pages.

Step 3: Connect the Tools

  1. Install Obsidian: Create a "Vault" pointing to your my-knowledge-base folder.
  2. Install Claude Code: Point the CLI at the same folder.
  3. Fill the raw/ folder: Dump everything—exported Apple Notes, saved PDFs, web clippings—into the raw/ directory. Don't worry about naming or organizing them.

🚀 Daily Workflows: How to actually use it

📥 Ingesting New Knowledge

When you find a great article or finish a book, drop the file into raw/ and run this command:

"I just added a new source to /raw. Read it, extract the key ideas, write a summary page in /wiki/summaries, update the INDEX.md, and link it to any existing concept pages."

🔍 Querying Your Brain

Instead of searching for keywords, ask Claude to synthesize:

"Based on everything in my wiki, what are the biggest gaps in my understanding of [Topic]? Compare what Source A says vs Source B."

Pro Tip: If Claude gives you a brilliant analysis, tell it: "This is a great insight. Save this as a new page in /wiki/analyses so I don't lose it."

🧹 The "Health Check" (Weekly Linting)

To prevent "AI hallucinations" from compounding, run a monthly health check:

"Review the entire /wiki/ directory. Flag any contradictions between pages, find orphan pages with no links, and suggest 3 new articles that would fill existing knowledge gaps."


🎯 Final Thought: Stop Shopping for Tools

The "Notion Trap" is spending more time configuring a workspace than actually thinking. The magic of this system is that the maintenance cost is near zero because the LLM does the bookkeeping.

The human's job is to curate sources and ask the right questions. Claude's job is everything else.

Stop searching for the perfect app. Create three folders, write one schema file, and start compounding your intelligence today.