An AI second brain with Claude Code and Obsidian: how I built and use it

I built a second brain, and I use it all day. My memory no longer lives only in my head, but in a system that remembers and works with me, for my business and beyond.

In this piece I'll show you how I built it, what it does for me every day, and how you can set one up. Call it AI in Obsidian, a second brain, or a searchable note app with an AI assistant on top: it comes down to the same thing. No theory from a whitepaper, just my own setup.

That system has two parts. The first is Obsidian, an app that stores your notes as plain text. My whole pile of notes (in Obsidian that's called a vault) is the memory: every client note, every meeting summary, every idea and every decision lives there, searchable. The second is Claude Code, Anthropic's AI assistant that runs on my own machine. Those are the hands: it reads that vault, writes in it, and gets work done with it. Not a stray chat window that forgets tomorrow what we discussed today, but an agent with permanent memory.

For me this isn't a luxury. I work solo, so there's nobody in the room saying: finish this first, you'll thank yourself later. So I built that voice myself. I have ADHD: without a system that catches what I drop, half of it hits the floor. The second brain is that partner. It catches my loose ideas before they vanish, and it holds me to what I planned with a clear head. Trust yesterday's you.

The gap nobody names: memory

Most people use AI like a slot machine. One prompt, one answer, and tomorrow you start from zero again. The real problem with tools like ChatGPT isn't the intelligence, it's the lack of memory: the context expires the moment you close the tab.

You don't need a complicated technical setup for this (developers call that RAG). A vault the agent can read and write to is enough. Obsidian is plain text, local, mine. Not locked inside one vendor's memory. If a better model shows up next year, my brain just moves with it.

Nothing gets lost here, and my AI can always reach it.

But don't ChatGPT and Claude remember things?

True, and it's a fine first step. ChatGPT remembers things across chats, and Claude does now too. For one-off tasks that really helps. But the moment you get serious, you run into three problems:

• You don't know exactly what it remembers. The memory is a black box the vendor controls.
• It's tied to one model. Switch to a better model or a different tool, and your memory stays behind.
• It isn't yours and it isn't structured. You can't search it, can't version it, and your own agents can't work in it.

Your own vault flips that around: you see every line, it travels with you to every model, and your agents read and write inside it. You start with built-in memory, but you end up with a second brain.

How I built my second brain

It sounds big, but it's a handful of parts working together. Let me walk through them.

The vault is the foundation. One folder of plain text files in Obsidian. Notes on clients, meeting summaries, ideas, decisions, daily logs. No database, no lock-in. Just folders and markdown I can still open ten years from now.

Claude Code is the engine. The CLI runs in that folder and gets to read and write. It knows how I work because I've written that down in an instruction file (my CLAUDE.md): who I am, how I write, what's fair game and what isn't. That file is the constitution. Every session starts there.

A memory layer on top. A few memory files the agent keeps up to date, so it remembers across sessions what we decided earlier. Not a vendor's black box, but files I can read and correct myself.

A search layer across it. A local search engine (qmd) over the whole vault, on keyword and on meaning. That means I can ask my entire business a question instead of digging through folders.

Hands reaching out. Connections via MCP, an open standard for hooking AI up to other software, to the tools I use every day: my calendar, my mail and MacWhisper for transcription. That way the agent can not only think but also do.

The brain grows on its own. A big chunk of the vault fills itself. Mail, calendar events and transcribed conversations land in the vault as notes automatically, without me retyping a thing. The more I work, the richer the memory gets, without extra effort.

A second agent as reviewer. Before anything is final, a separate, critical agent checks the first one's work. That way I catch mistakes and inconsistencies before they leave the building.

A second brain isn't an archive, it's a garden

A vault you never tidy up silts shut. That's why maintenance is part of it, and I have a few fixed rituals for that. At the end of every workday one command captures everything from that day: what I did, what I decided, and which mistakes I nearly repeated. Every so often I prune, kick out duplicates and stale notes, so the memory stays sharp instead of clogging up. You don't need strict zettelkasten discipline; a bit of tidying is enough. A second brain you don't maintain becomes the same junk drawer as the systems you wanted to replace.

Everything stays local

This is the part that weighs heaviest for small and mid-sized businesses. My vault runs 100% locally. It sits as plain text on my own machine, not in a vendor's cloud. Private data and client information don't leave my computer unless I explicitly allow it. When Claude Code uses a file, that specific content goes to Anthropic's cloud model, and I make that call deliberately every time. The rest stays local.

And it can go stricter still. If I want nothing to go out at all, I run a local AI model on my own machine and everything stays in-house. In practice I work hybrid. On my end that runs on a Mac Studio: a local model handles the search layer and the simple chores like summarising and classifying, and only the work that needs nuance goes to the heavier Opus model in Claude Code. So I decide what stays local and what gets to reach a stronger model.

What has to stay private never leaves my machine.

The recipe

You don't need to be a developer to do something with this. Here's the shortest route, from small to complete.

What you need:

• Obsidian, or really any folder of markdown files
• Claude Code, the CLI
• A CLAUDE.md with your context and your rules
• A transcription tool like MacWhisper that writes conversations away automatically
• Optionally a local search layer like qmd for the "ask your whole business" trick
• A handful of fixed commands for recurring jobs

The steps:

1. Make your vault. One folder, a few subfolders (clients, meetings, ideas, decisions, daily). Start small.
2. Set up Claude Code in that folder and write your CLAUDE.md. Who you are, how you work, what the agent may and may not do. This is the most important piece.
3. Add a memory layer, so the agent remembers what's in play across sessions.
4. Hook up transcription, so conversations land in your vault without retyping.
5. Put a search layer over the vault for the "ask your whole business" trick.
6. Build your first command. Have it neatly save everything at the end of the day.
7. Connect the tools you work with, like your calendar and mail.
8. Keep improving. Any job you do more than twice, turn into a command.

What it does day to day

Four things I use it for every day.

A conversation turns itself into knowledge. I record a client call. MacWhisper turns it into text automatically, and an agent pulls out the summary, decisions and action points and files them into my vault. I type nothing. A week later I still know exactly what we agreed, because it sits searchable in my brain.

A week of input becomes content. One command reads the videos and articles I saved that week and distills the useful insights out of them. A second turns those insights into concrete content ideas. What I consume becomes raw material for what I make, automatically.

I run my business through it. My calendar, my mail and my social posts all run through that same brain. The agent proposes, I approve or adjust. I almost never type a reply from scratch. That's the difference between a chatbot and an AI agent that takes work off your hands.

I ask my whole business something. "What did we agree with that client back in March?" "What's the status of that proposal?" Instead of digging through folders and inboxes, I ask the question and the answer comes back from everything I've ever written down. My business has become searchable.

Own commands: from memory to insight

Remembering and retrieving information is only half the work. The real win is in synthesis: a brain that draws connections I'd never have written down myself. For that I wrote a handful of my own commands, short instructions I type into Claude Code that think through my entire vault in one go.

Here are five I use nearly every day. For each one, I'll show what it produces.

/emerge surfaces conclusions my notes imply but never spell out. My most valuable command.

/morningops builds my morning briefing out of mail, calendar, vault and git, and puts it all in one note.

/challenge holds my assumptions up to the light and hunts down where my own notes contradict each other.

/drift compares my intentions with my actions over the past 60 days. That's how I see what I'm avoiding.

/closeday closes out my workday: it summarises what happened, pulls out the action points and draws connections I'd otherwise miss.

The point isn't that these are off-the-shelf features someone else designed for me. I wrote them myself, in plain language, exactly the way I work. The brain bends to me, not the other way around.

These five are a taste of a set of twelve. I cleaned them up, cut them loose from my own vault and put them on GitHub, so you can download and use them yourself: the twelve commands are open source on GitHub. They run in any AI coding agent (Claude Code, Cursor, Codex) on any folder of markdown notes, and they're read-only: they read your notes and think along, but write nothing back. Clone the repo, run the install script and point them at your own vault.

DataDreamOS: my business runs on it

The vault started as memory, but it's grown into the operating system of my business. I call it DataDreamOS. My whole back office lives in it, and the agents work with it directly.

A few examples from the trenches:

• I generate invoices from the data already in the vault, not in some separate bookkeeping app I have to fill in by hand.
• I keep my finances, leads and sales pipeline in there: who's in which stage, what's outstanding, what's coming up.
• Per client and contact I have one overview: every conversation, every meeting and every open action in one place.
• Action points and tasks come out of my conversations and notes automatically, so nothing slips through.

They're not separate tools sitting next to each other, they're one system where everything comes together. That's the difference between loose AI tricks and a business that actually runs on AI.

This is agentic engineering, not vibe coding

Agentic engineering means you deploy AI like an engineer would: with structure and control, not on a hunch. This is where the difference that matters shows up. Refuse AI and you fall behind. But let AI loose without discipline and you ship brittle work that collapses at the first edge case. The sweet spot sits in between: the discipline of an engineer, executed through AI.

In practice that means structure first, execution second. I lay down up front what's allowed and what isn't, what the output has to look like, and which steps come in which order. The prompt isn't the work, the prompt is the result of a thought-through workflow. Only once that scaffolding is in place do I let the agent loose.

Vibe coding is typing a prompt and hoping. Agentic engineering is building a system you can tune, repeat and hand off.

Structure first, execution second.

What this means for you

A second brain sounds big, but the core is simple: your knowledge lives in one place and your AI can reach it. That's the difference between starting over every day and building on what you already knew yesterday.

You don't have to rebuild my setup. The principle is what counts: make sure your context is yours and sits in one searchable place, and work out your process before you point AI at it. A business that scatters its knowledge across six systems and throws AI on top gets loose tricks. A business that has its memory in order and its workflow drawn out gets a second brain that works with it.

And this isn't just for businesses. The same principle works just as well for your personal life: your plans, your notes, your admin, everything that now vanishes into scattered apps and inside your head.

And this is what it really comes down to. The AI model on its own doesn't give you an edge, everyone rents the same model for a few tenners a month. Your edge lives in what the model doesn't have: your context, your conversations, your way of working, in one place your AI can reach. That's yours, and nobody can replicate it overnight.

That's the difference between AI as a toy and AI as a colleague that never forgets. If you want to see what that looks like for your own processes, that's exactly the work I do: AI solutions that run in production.