Build Your AI-Powered Second Brain with Claude and Obsidian

The Core Problem: Storage vs. Retrieval

Your biological brain excels at thinking but fails at storing and organizing information reliably. Most people forget tasks, deadlines, and key details. A second brain solves this by acting as an external librarian that never forgets, always organizes, and makes information instantly available when you need it.

An 80-Year-Old Vision Finally Realized

Vannevar Bush conceived the Memex in 1945—a desk that recorded everything you read with trails connecting related ideas. For decades this remained theoretical because no system could maintain such a database automatically. Large language models now make this possible: they can read, file, interconnect, and update thousands of documents on autopilot.

The second brain concept evolved from analog dream to AI-powered reality

The Three-Layer Architecture

Obsidian is the IDE (development environment where you shape ideas), Claude is the programmer (manipulates and builds structure), and the wiki is the codebase (the growing library of interconnected pages you query). Together they form a feedback loop: you feed raw data in, the AI organizes and links it, and you extract insights out.

Three-layer system for building a second brain

Obsidian: Local-First, Markdown-Based Notes

Obsidian stores all notes as plain markdown files on your computer—not in the cloud. Markdown is simple: one hashtag for headings, asterisks for bold/italic, double brackets for cross-links. If Obsidian disappears tomorrow, your files remain yours forever. The graph view visualizes all connections as an interactive network of nodes.

Cross-Linking: From Pages to Networks

In Obsidian, you link documents using double brackets: [[OpenAI]]. Each link creates a connection in the knowledge graph. As you add more pages and links, isolated documents become a web. The graph view animates this transformation, showing how topics cluster and relate. This is what turns a note-taking app into a knowledge network.

Claude Code Desktop: The AI Librarian

Claude Code Desktop is Anthropic's unified interface combining a code editor, browser, and chat with Claude. You can ask it to build entire folder structures, ingest URLs, update documents, and run automations. It includes a built-in browser and remote-control capability (via /remote control) so you can control it from your phone.

Why Both Tools Matter

Obsidian handles storage and visualization; Claude handles ingestion, organization, and analysis. Obsidian keeps you independent of any vendor (files are yours). Claude provides the AI labor to maintain the system 24/7 without human effort. Together they create a system that compounds in value over time.

The Flat Folder Structure

Create three top-level folders: Inbox (quick captures), Raw (immutable source documents), and Wiki (organized knowledge). Inside Wiki, add subfolders for Concepts, Entities, Summaries, Templates, and Skills. Keep the structure flat—avoid deep nesting because it confuses LLMs. Topics live in links, not folder hierarchies.

Recommended vault folder structure

The Rulebook: claude.md

Create a claude.md file that serves as an employee rulebook. Every morning Claude reads it and knows its job: raw files are immutable sources, wiki pages are compounding knowledge, inbox items await processing. This file defines how Claude should behave, what to prioritize, and how to maintain the system. You can ask Claude to write this for you.

Optional Plugins: DataView and Kanban

DataView creates automatic tables and queries across your notes. Kanban adds a visual board where you drag tasks through columns (To Do → Doing → Done). Both are community plugins in Obsidian settings. They're optional on day one but powerful once your vault grows.

Setup Timeline

You can build a functional second brain in a single afternoon. Create the folder structure, write or have Claude write the rulebook, add a few initial documents, and start ingesting data. The system compounds over time—after 10 ingestions, isolated dots become a web.

1 afternoon

Time to build a functional second brain

Initial setup is fast; value compounds over weeks and months

Ingesting 22,000 Posts for Under $100

Wes built an X (Twitter) data ingestion engine that archives 22,000 posts representing 5.9 billion views from 4,400 unique authors. The entire system cost under $100 in API credits. It runs offline—no API calls needed to search. The system automatically computes trending velocity, identifies algorithmic changes, and benchmarks performance against other accounts.

22,000 posts

Archived in second brain

5.9 billion views, 4,400 authors, under $100 cost

Detecting Algorithmic Shifts

By analyzing tweet performance over time, the system detected that the X algorithm changed a few months ago. Impressions that had been climbing month-over-month suddenly dropped. The system identified which tweet formats (standalone video, quote + video, quote + image, etc.) still work and which no longer perform, allowing Wes to adjust strategy before losing more reach.

System alerts when platform behavior shifts

From Raw Data to Actionable Playbook

The system ingests raw X data, organizes it into weekly scorecards and performance metrics, then generates a living document called the X Growth Playbook. This isn't about growth hacks—it's about understanding how the algorithm works so you don't get caught off-guard by changes. Claude updates these insights daily.

Kanban Board for Video Pipeline

Create a Kanban board with columns: Idea → Packaging Gate → Research → Scripting → Filming → Editing → Published. As ideas move through stages, Claude can automatically generate suggested hooks, research summaries, and fact-check sheets. For example, when a video idea moves to Research, Claude pulls the transcript from a related 3-year-old video and identifies what's changed since then.

Video production pipeline with AI automation at each stage

AI-Generated Hooks and Research

When you move a video idea to Research, Claude automatically generates suggested hooks based on trending data and your past videos. Example hook: 'Three years ago, I showed you a dad selling Excel formulas for $25,000 a month. The number today made me double-check my sources.' Claude also identifies what happened to past case studies (e.g., thumbnailest.com sold for six figures in 2024) and warns about platform risks.

Sponsor Tracking Kanban

A separate Kanban board tracks sponsor deliverables: Script → Approval → Recording → Editing → Publishing. As you drag items across columns, the system updates status. This can trigger notifications (email, SMS, Slack) if you fall behind. For people with ADHD or executive function challenges, this replaces the 'ADHD tax' of forgotten deadlines.

Three Layers of Output: War Room, Wiki, Doctrine

The War Room (a 24/7 mini PC) gathers raw data from YouTube, X, news platforms. The Wiki organizes and cross-links everything. The Doctrine is the output layer: Claude (using Claude 3.5 Sonnet for deep analysis) generates actionable strategies backed by actual data. Every claim in the Doctrine traces back to collected evidence.

Three-layer system for intelligence gathering and strategy

The Armory: Future Projects and Ideas

The Armory is a folder of projects Claude suggests you should build. Examples: X data ingestion engine, analytics dashboard, comment archive, clip engine, retention miner. Claude ranks these by feasibility and impact. You choose which to prioritize. This turns your second brain into a product roadmap.

The OODA Loop: Observe, Orient, Decide, Act

The system runs a continuous feedback loop: Observe (data collection via War Room), Orient (wiki summaries and doctrine), Decide (strategy generation), Act (execute and collect results). As time passes, the loop compounds—better models improve analysis, more data improves decisions, and execution results feed back in.

Continuous OODA loop for compounding improvement

Compounding Returns Over Time

As the system runs longer, it improves in two ways: (1) more data accumulates, making analysis richer, and (2) better models are released, making analysis smarter. Even if Claude is replaced by a better model, the structured data and insights remain useful. The system doesn't depend on any single vendor.

You Own Your Data Forever

All files are stored locally as plain text markdown. No vendor lock-in. If Obsidian shuts down, your files remain. If you switch from Claude to another model, the wiki structure stays intact. You're not dependent on any company's API or service continuing to exist.

The System Improves as Models Improve

Better LLMs make better summaries, better cross-links, better insights. The system doesn't decay or become obsolete—it gets better with each new model release. Your accumulated data becomes more valuable as AI capabilities increase.

Three-Step Setup Process

Step 1: Download Obsidian (free) and Claude Code Desktop (free to start). Step 2: Create vault with three folders (Inbox, Raw, Wiki) and write or have Claude write the rulebook. Step 3: Start ingesting data—one link a day, or set up automation to pull data you care about. After 10 ingestions, dots become a web.

Three-step setup for your second brain

This Is Not Optional Learning

Building a second brain requires learning new tools and workflows. Learning is uncomfortable—there's resistance. But push through it. Once in place, the system compounds. Wes wishes he'd built this on day one when Karpathy first proposed it. The earlier you start, the more value you capture.