How to Build a Persistent Second Brain with Claude-Obsidian in 2026?

Build a Persistent Second Brain with Claude-Obsidian: Tired of your AI forgetting everything the second you close the chat? Or staring at an Obsidian vault full of unread notes that never compound into real knowledge?

There’s a powerful solution: claude-obsidian, a free, open-source Claude Code skill that turns your Obsidian vault into a self-maintaining, compounding wiki based on Andrej Karpathy’s LLM Wiki pattern.

Every PDF, article, transcript, or conversation you feed it gets automatically extracted into structured wiki pages with frontmatter, wikilinks, citations, and contradiction flags. Your knowledge grows like compound interest—smarter, more connected, and never lost.

This complete setup guide (updated for 2026) walks you through everything step-by-step. By the end, you’ll have a true second brain that Claude never forgets, because it runs directly on your local markdown files.

Why Claude-Obsidian Beats Traditional Obsidian + AI Tools

Most Obsidian AI plugins or RAG setups just let you “chat over your notes.” You still do all the linking, tagging, and upkeep.claude-obsidian is different—it’s inspired by Karpathy’s vision of an LLM as a tireless wiki maintainer.

• Drop in raw sources (PDFs, URLs, transcripts, meeting notes).
• Claude auto-creates 8–15 interconnected wiki pages per source.
• It handles entities, concepts, cross-references, and flags contradictions.
• A “hot cache” keeps recent context loaded for every session.
• Your vault becomes a living knowledge graph that gets smarter daily.

Result after a month? A breathtaking Obsidian graph view full of clusters, connections, and insights—no manual labour required.

Prerequisites (5 Minutes to Gather)

1. Obsidian (free desktop app) → Download here
2. Node.js (v18+ LTS recommended)
3. Claude Code (Anthropic’s official CLI agentic coding tool) → Install via npm install -g @anthropic-ai/claude-code (requires a Claude subscription for full power)

That’s it—no extra paid tools or complex setups.

Step-by-Step Setup: Install claude-obsidian

Option 1: Easiest (Recommended) – Clone as a Ready Vault

bash

git clone https://github.com/AgriciDaniel/claude-obsidian
cd claude-obsidian
bash bin/setup-vault.sh

Open the folder as a vault in Obsidian. Then open Claude Code in the same folder and type /wiki.

Option 2: Install as a Claude Code Plugin (for existing vaults)

In Claude Code:

bash

claude plugin marketplace add AgriciDaniel/claude-obsidian
claude plugin install claude-obsidian@claude-obsidian-marketplace

Then run /wiki to bootstrap.

Claude will automatically create the folder structure, initialize the index, hot cache, and templates.

The 3 Core Commands That Run Your Second Brain

Once set up, everything happens through these simple slash commands in Claude Code:

• /save → After any valuable conversation, type this. Claude reads the full thread, extracts key ideas, creates properly formatted wiki pages with wikilinks, and updates the index. Conversations never disappear again.
• /autoresearch [topic] → Need deep research? Claude runs 3–5 rounds of web research, synthesizes findings, adds citations, and files everything into your wiki. Customize priorities in skills/autoresearch/references/program.md (e.g., “prefer PubMed” for medical topics).
• /canvas → Turn knowledge into visuals. Claude generates an infinite Obsidian Canvas with nodes, flowcharts, timelines, or knowledge graphs. Drop images, PDFs, or notes directly onto it.

The Daily Ingest Loop (The Magic Habit)

1. Drop anything worth keeping into the .raw/ folder (PDFs, articles, video transcripts, URLs—nothing gets modified here).
2. In Claude Code, type: ingest all of these (or specify files).
3. Claude processes it: creates interconnected pages, adds wikilinks (average 12 per page), flags contradictions with ![!contradiction] callouts, and logs everything in wiki/log.md.

Ask anything anytime: Claude scans the hot cache + index and cites exact pages in your vault (not its training data).

How It Stays Cheap and Scalable (Even With Thousands of Pages)

No expensive full-vault RAG here. The system only loads:

• hot.md (~500 tokens of recent context)
• index.md (one-line summaries)
• Only the relevant wiki pages for your query

Token usage barely increases as your vault grows. Perfect for power users.

What Your Vault Looks Like After 30 Days

• Beautiful, color-coded graph view showing concept clusters.
• Automatic backlinks and cross-references everywhere.
• A living knowledge base that feels like Jarvis from Iron Man.

Users report feeling like “Tony Stark with Jarvis” after just a week.

Pro Tips for Maximum Results

• Use the built-in lint command regularly: lint the wiki (fixes orphans, dead links, etc.).
• Enable recommended Obsidian plugins (Bases, Graph View, Canvas, etc.)—the setup script handles most.
• For visual thinkers, combine with /canvas commands for instant mind maps and presentations.

Ready to Build Your AI Second Brain? claude-obsidian is 100% free, open-source (MIT license), and actively maintained on GitHub: github.com/AgriciDaniel/claude-obsidian.

Stop losing knowledge. Start compounding it. Install it today and watch your personal knowledge base evolve into something truly intelligent. What will you ingest first? Drop your experience in the comments—I’d love to hear how your graph view looks after a month!

This guide is based on the official claude-obsidian repository and Karpathy’s original LLM Wiki concept. Always check the repo for the latest updates.