A growing number of people are building their own second brain by connecting an AI tool to a notes app. The dominant setup pairs Claude Code with an Obsidian vault: Claude reads and writes markdown files directly, creating links and organizing notes automatically. The open-source claude-obsidian project alone has over 9,000 GitHub stars.
The trend traces back to Andrej Karpathy’s LLM Wiki pattern: use an LLM to maintain a wiki by feeding it raw information and letting it do the structuring. People took that idea and applied it to their personal knowledge bases.
If you’re not sure what a second brain app is, our category guide covers the basics. And if you’re comparing second brain apps more broadly, our roundup breaks down the options.
Here’s what the trend looks like in practice, why it’s popular, and where it falls short.
What the setup actually looks like
The core is deceptively simple. Obsidian stores notes as plain markdown files on your local disk. Claude Code can read and write those files directly. No API, no plugins, no export steps.
cd ~/second-brain
claudeThat’s it. Claude now has full read/write access to every note in your vault.
The magic happens through a file called CLAUDE.md at the vault root. This is Claude’s instruction manual — it tells every new session how the knowledge base is structured, what conventions to follow, and what’s available. Without it, you’d spend the first five minutes of every session explaining your system. With it, Claude already has the full context and can follow your instructions immediately.
A typical vault structure looks like this:
~/second-brain/
├── CLAUDE.md ← Claude's persistent context
├── daily/ ← daily notes
├── projects/ ← one folder per active project
├── areas/ ← ongoing responsibilities
├── resources/ ← reference material, clippings
├── _inbox/ ← unprocessed capture
└── templates/ ← note templatesPeople run Claude Code inside Obsidian using one of three methods: a terminal plugin that opens a real terminal inside Obsidian, an MCP bridge that exposes vault operations as structured tools, or a sidebar embed that puts Claude directly in Obsidian’s right panel.
Real workflows people use
The setups support several documented workflows:
Daily notes. Each morning, you ask Claude to create today’s daily note, carry over incomplete tasks from yesterday, and surface active priorities from your projects.
Research and capture. You ask Claude to research a topic, search the vault for existing notes, synthesize findings into a structured note, and link it to related content already in the vault.
Meeting notes to action items. Claude reads a meeting note, extracts tasks assigned to you, and creates structured task items linked back to the original note.
Weekly review. Claude reads all daily notes from the week, extracts completed and incomplete tasks, identifies patterns, and generates a summary with next week’s priorities.
Connecting the dots. Claude reads all notes on a topic from the last few months, finds recurring themes you haven’t explicitly connected, and creates a map of content.
These workflows are real and they work. The setup is more capable than most people assume.
Why people do this
Claude forgets between sessions. Every time you start a new Claude Code session, you’re starting from scratch. The Obsidian vault is Claude’s persistent memory — it grows with every session and carries context forward.
You already have the notes. Most people have years of accumulated notes, meeting transcripts, research, and ideas. Migrating to a new app means losing that context. Connecting AI to what you already have is faster than starting over.
You own the data. Obsidian stores plain markdown files. You control the underlying data, and you can move it anywhere later. There’s no vendor lock-in at the data level.
The schema file is powerful. A well-written CLAUDE.md turns Claude from a general-purpose assistant into a domain-specific knowledge worker. You define the structure, the conventions, and the quality bar.
It’s a proof of concept. The DIY approach proves that AI can turn a passive notes app into an active second brain. It shows what’s possible when AI understands your information.
Where the setup falls short
The DIY approach works until you push it beyond what a folder of markdown files can handle. Here’s what breaks down:
No real-time capture. These setups process existing notes. They don’t help you capture new information as it comes in. When you’re in a meeting, on a call, or have a quick thought on your commute, you still need a separate app to record it before Claude can process it later. The gap between capture and processing is where information gets lost.
No structured output. Claude writes text into files. It doesn’t create separate entities like tasks, contacts, or knowledge entries. Action items live inside text notes, not in a task board you can filter and sort. People you mention stay as text, not as profiles with interaction history. You get a wiki, not a system.
No proactive surface. Claude processes your notes when you run it. It doesn’t check in on stale tasks, remind you about follow-ups, or tell you what to focus on today. You have to prompt it to look for something specific. The system is reactive, not proactive. As we discuss in how to stop forgetting follow-ups, the gap between “I noted this” and “I actually followed through” is where most commitments die.
No dedicated UI. Everything happens through chat. You type a prompt, Claude responds with text. There’s no task board to scan at a glance, no calendar view for deadlines, no contact list you can browse. Even simple actions — checking what’s due today, seeing which tasks are stale, reviewing your week — require typing a prompt and reading a text response back. A dedicated interface gives you instant situational awareness. Chat forces you to ask for it every time.
No mobile experience. The setup lives in your terminal and Obsidian desktop app. There’s no way to capture a voice note on your commute, check your task board from your phone, or quickly log a thought before it slips. As we cover in how to stay focused when your day is interrupted, the ability to capture without switching tasks is what protects deep work.
What the trend reveals
The DIY AI + notes trend reveals two things:
1. People want AI in their second brain. The demand is real. People are willing to build technical workarounds because they want their knowledge base to be active, not passive. They want AI that understands their information and helps them act on it.
2. Notes apps aren’t enough. The fact that people are building these setups shows that storing notes is no longer sufficient. They want AI to extract tasks, connect related information, surface follow-ups, and maintain structure automatically. A notes app with search doesn’t satisfy that need.
If you’re already doing it
If you’ve built a DIY setup and it works for you, keep using it. The workflows are real and the knowledge compounding is genuine. Each new source ingested becomes more valuable because it connects to everything already in the vault.
But if you find yourself spending more time maintaining the system than using it, or if you’re hitting context limits, or if you want real-time capture and proactive reminders without building them yourself, it’s worth considering a purpose-built alternative. The DIY approach is a stepping stone, not a destination.
A purpose-built second brain app removes the bridge between AI and your knowledge base. Capture is fast because the system is built for it. Structured output is native because the app understands what a task is, what a contact is, and what a knowledge entry is. Proactive nudges work because the system tracks deadlines and follow-ups as entities, not as text inside notes. And it works on mobile, so you can capture on the go.
The DIY trend proves the problem is real. A purpose-built app proves the solution doesn’t need to be a hack.