Build an AI Second Brain with Claude + Obsidian — Karpathy's LLM Wiki Method (Full Guide)

Build an AI Second Brain with Claude + Obsidian — Karpathy's LLM Wiki Method (Full Guide)

Building Your Second Brain

Introduction to the Concept

  • The speaker introduces the idea of a "second brain," emphasizing how current notes and ideas are scattered across various platforms, leading to inefficiency.
  • A second brain is contrasted with chat history; it compounds knowledge rather than just storing it, allowing for better recall and connection of ideas.

The Setup Process

  • The setup process is described as taking one evening, referencing Andrej Karpathy's work on building LLM knowledge bases in 2026.
  • The speaker outlines three layers essential for creating a second brain: storage (Layer 1), the brain (Layer 2), and autopilot (Layer 3).

Layer One: Storage

  • Obsidian is introduced as the tool for Layer 1, where all notes are stored as plain text files on your computer, ensuring ownership and longevity.
  • Notes can be linked together in Obsidian, forming a graph that visually represents how ideas connect over time.

Creating Your Vault

  • Instructions are provided to create a new vault in Obsidian, which will serve as the foundation for your second brain.
  • Users need to install Claude from claude.com/download and sign up for a paid plan to access advanced features.

Connecting Claude to Your Vault

  • Steps are outlined to connect Claude with Obsidian using community plugins and API keys.
  • A common mistake regarding API key usage is highlighted; users should exclude the word "bearer" when entering their key.

Initializing Your Second Brain

  • Users are encouraged to have Claude interview them to populate their initial notes automatically instead of typing everything manually.
  • Projects should be organized into folders within the vault, allowing Claude to focus on specific tasks without overwhelming information.

Skills and Autopilot Features

  • Skills represent saved workflows that Claude can execute based on user commands, streamlining repetitive tasks.
  • Integration with Google Calendar and other tools allows real-time updates and task management through simple commands.

Automating Maintenance Tasks

  • Users can set up routines within Claude’s desktop app that automate filing new information while they sleep.
  • Emphasis is placed on granting read-only access wherever possible to maintain control over data integrity.

Final Thoughts on Building Trust with AI Agents

  • The importance of controlling AI agents through access permissions rather than prompts is discussed as crucial for safety.

Community Resources

Open Source Solutions

  • Several open-source projects related to building a second brain using Claude are mentioned, including options tailored for different user needs like executive or creator roles.

Conclusion: Growing an Asset

Long-term Benefits

  • The speaker concludes by highlighting that unlike traditional chatbots which start from scratch each time, a second brain accumulates knowledge over time, becoming an invaluable asset throughout one's career.
Video description

Your best ideas are scattered across notes apps, browser tabs, and Claude chats you'll never find again. A second brain fixes that: one place that holds everything you know, with Claude reading it, linking it, and getting smarter every single day. This is the full build — Claude + Obsidian — based on Andrej Karpathy's "LLM Wiki" pattern. Every click and command spelled out, even if you've never opened Claude Code or Obsidian. We build it in 3 layers + 1 rule: • Layer 1 — STORAGE: a plain-text Obsidian vault you own forever • Layer 2 — THE BRAIN: Claude Code reading, linking & growing it (MCP) • Layer 3 — AUTOPILOT: it files itself overnight with Routines • THE RULE: keys, not prompts (the one thing you never break) ⏱️ Tip: this is a dense walkthrough — watch at 1.25–1.5x if you like a faster pace. 📘 FREE — The Second Brain Build Checklist (every command + every copy-paste prompt): Comment WIKI and we'll send it, or grab it here → https://hyperautomationlabs.co/free/second-brain 🔔 We go this deep every week — subscribe + follow: • YouTube: https://youtube.com/@hyperautomationlabs1045 • Instagram: https://instagram.com/hyperautomationlabs • Facebook: https://facebook.com/HyperautomationLabs 🚀 Want to go pro? Our paid guides go deeper: • Claude Code Guide • OpenAI Codex Guide • Claude / AI Sales Guide • CCA-F Certification Prep Kit → https://hyperautomationlabs.gumroad.com 🔗 Tools & repos shown (all verified): • Obsidian — https://obsidian.md • Claude — https://claude.com/download • Obsidian Local REST API — https://github.com/coddingtonbear/obsidian-local-rest-api • mcp-obsidian — https://github.com/MarkusPfundstein/mcp-obsidian • workspace-mcp — https://github.com/taylorwilsdon/google_workspace_mcp • Karpathy's "LLM Wiki" gist — https://gist.github.com/karpathy Ready-made second brains: AgriciDaniel/claude-obsidian · eugeniughelbur/obsidian-second-brain · coleam00/second-brain-starter ⏱️ Chapters 0:00 Your ideas are scattered 0:30 Why a second brain compounds 0:54 Where this came from: Karpathy's LLM Wiki 1:26 The map: 3 layers + 1 rule 1:49 Layer 1 — Storage (Obsidian) 2:21 Make your vault + the [[link]] mechanic 2:50 Layer 2 — The Brain (Claude Code) 3:19 Connect Claude to your vault (MCP) 4:08 Load yourself in (CLAUDE.md) 4:39 Projects with the right structure 5:21 Build a skill (stop repeating yourself) 5:52 Layer 3 — Autopilot (live data) 6:32 Put it on Routines (7am autopilot) 7:02 The one rule: keys, not prompts 7:33 Skip the setup: 3 ready-made repos 8:15 Why it compounds 8:39 Free checklist + go deeper #ClaudeCode #Obsidian #SecondBrain #AI #Karpathy #LLMWiki #Anthropic #PKM #Productivity