A Second Brain That Organizes Itself While You Work
A complete Obsidian and Claude Code setup for turning articles, transcripts, and raw notes into searchable knowledge.
second brain playbook
Build Your Second AI Brain
The Complete Guide - Obsidian + Claude Code
STEP 1 - Copy Karpathy's idea file from this link:
https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
You'll paste this into Claude Code during setup. Keep it open.
How It Works
Your articles, Organized
Claude Code Ask anything,
notes & fi fi knowledge fi
organizes it all get cited answers
transcripts base in Obsidian
n Keep adding sources - it gets smarter every time
Setup (5 Minutes)
1. Download Obsidian
Go to obsidian.md - free. Create a new vault, give it a name.
2. Open Claude Code in that folder
Terminal: cd into the vault folder, run claude.
3. Paste the blueprint
Paste Karpathy's idea file (from the link above), then add this underneath:
You are now my LLM wiki agent. Implement this idea file as my complete second brain. Create the CLAUDE.md schema, the folder structure, the index, the log. Set everything up so I can start dropping in sources.
Creates: raw/ (your sources) + wiki/ (AI-built pages) + CLAUDE.md + index.md + log.md
4. Install Web Clipper
Chrome extension: 'Obsidian Web Clipper.' In settings, change default folder to raw.
5. Add your first source and ingest
I just added [article name] to the raw folder. Please ingest it into the wiki.
6. Query it
Based on everything in the wiki, [your question]. Cite specific wiki pages.
What A Wiki Page Actually Looks Like
When you ingest an article, Claude Code creates pages like this:
# OpenAI Type: Organization Founded: 2015 Key People: [[Sam Altman]], [[Greg Brockman]], [[Ilya Sutskever]] ## Summary Leading AI research lab originally founded as a nonprofit. Developed GPT series and ChatGPT. Transitioned to capped-profit model in 2019. ## Key Connections - Competitor to [[Anthropic]] and [[Google DeepMind]] - Central player in [[AI Safety]] debates - Driving force behind [[Compute Scaling]] trends ## Mentioned In - [[AI 2027]] - referenced as primary AGI developer - [[Core Views on AI Safety]] - compared to Anthropic approach ## Open Questions - How will open-source strategy evolve? - Relationship with Microsoft post-restructure?
Every [[link]] connects to another page. That's how the graph builds itself.
Three Upgrades
Upgrade 1: Hot Cache
Your agents crawling 100 pages for every question is slow and expensive. Fix: one tiny file they read first.
Add a hot cache. Create hot.md in wiki/ - a ~500 word summary of the most important context. Update it automatically on every future ingest.
Upgrade 2: Lint & Health Check
Wikis get messy over time. Pages contradict each other, concepts go missing. Run this weekly:
Run a health check. Find contradictions, orphan pages, missing concepts, and gaps where a web search could help. Suggest fixes.
Upgrade 3: Agent Integration
Any AI that reads files can read your wiki. Add this to the CLAUDE.md in your other projects:
When you need context about me or my work: 1. Read [wiki-path]/wiki/hot.md first 2. If more detail needed, read wiki/index.md 3. Follow links to relevant pages 4. Only read from wiki/ - never raw/
What You Can Actually Do With This
After a strategy meeting: Drop in the transcript. Next week ask: 'What did we commit to and what's still unresolved?' - cited answer in 10 seconds. No re-reading the transcript. Planning content: Ingest 30 of your own video transcripts. Ask: 'What topics have I covered most? What frameworks do I keep referencing? What gaps haven't I filled?' - pattern recognition across your entire body of work. Researching a new field: Clip 10 articles over two weeks. By article 5, the wiki is already cross-referencing concepts for you. Ask: 'What are the competing views on X and which sources support each side?' - with citations. Running a business: Ingest your ops docs, tool stack, team structure, and strategy. Ask: 'Based on my distribution architecture and current AI trends, where's my biggest growth opportunity?' - it connects your data to the research. Preparing for a meeting: Ingest all relevant docs the night before. Morning of, ask: 'Give me a 3-minute briefing on the key issues and what I should push for.' - instant prep from everything you've filed.
Keeping It Alive
Daily: Clip interesting articles into raw/. 5 seconds each. Ingest when you have a minute. Weekly: Run the lint health check. Fix what it flags. Monthly: Review graph view. Which clusters are growing? Which are stale? Ask: 'What are the biggest gaps in this wiki right now?' Always: File good query answers back as wiki pages. Your best thinking should compound, not disappear into chat history.
Quick-Start Checklist
n Download Obsidian (obsidian.md) - create a vault n Open Claude Code in the vault folder n Paste Karpathy's idea file + kickoff prompt n Install Obsidian Web Clipper - default folder to 'raw' n Ingest first source - check graph view n Add second source - watch cross-connections n Set up hot cache (Upgrade 1) n Run first lint check (Upgrade 2) n Point other agents at the wiki (Upgrade 3) Karpathy's idea file: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f Full video: youtube.com/@AIEdge | Free community: Telegram link in video description