Why I Ditched Notion for Obsidian — And Let My AI Agents Into My Second Brain

How a simple folder structure replaced my $96/year subscription and made my AI assistants 10x more useful.

I've been a Tiago Forte Second Brain disciple for years. Notion was my PARA system. Every highlight, every meeting note, every fleeting idea — captured, organized, retrievable.

But here's what nobody tells you about Notion as a Second Brain: it's a walled garden that your AI tools can't touch.

When I started building AI agents to help run my portfolio of micro-businesses, I hit a wall. My agents could write code, send emails, manage deployments — but they couldn't access the one place where all my knowledge lived. My Second Brain was locked behind an API that required custom integrations for every single tool.

So I moved everything to Obsidian. And something unexpected happened.

The Setup That Changed Everything

My setup is dead simple:

Obsidian pointing at a shared folder (~/shared/)
Syncthing keeping that folder in sync across 3 machines (MacBook, Mac Mini, EC2 server)
QMD (semantic search) indexing everything for natural language queries
Readwise auto-syncing my Kindle highlights
PARA structure (Projects, Areas, Resources, Archive)

Total cost: $0 for the infrastructure. I already had a Readwise sub.

Here's the magic: my AI agents read and write to the exact same folder. When my agent Marvin creates an infrastructure SOP at 2am, I see it in Obsidian when I wake up. When I highlight a passage in a Kindle book about positioning strategy, my marketing agent Ace can search for it the next morning.

No API. No webhooks. No integration layer. Just files.

Why Notion Failed Me (And Might Be Failing You)

Don't get me wrong — Notion is beautiful software. But for a solopreneur running AI agents, it has three fatal flaws:

1. Your AI can't just... read it. Every AI tool needs a custom Notion integration. MCP servers, API tokens, database IDs. It's a project every time. With Obsidian on the filesystem, any tool that can read a text file can access your knowledge base. That's every AI tool ever built.

2. You don't own the search. Notion's search is Notion's search. You can't swap it, extend it, or run it locally. With QMD, I have semantic search that runs entirely on my Mac Mini's GPU — no data leaves my machine. I can search "what did we decide about email security?" and get relevant results from across hundreds of notes.

3. It doesn't scale to multiple agents. I have two AI agents (Marvin for infrastructure, Ace for marketing) plus my own brain. All three need to read and write to the same knowledge base, simultaneously, from different machines. Notion's collaboration is designed for humans in browsers, not agents on servers.

The PARA Structure (With a Twist)

I use Tiago Forte's PARA system, but with a twist for agent compatibility:

~/shared/
├── 0-Inbox/           → Capture & triage
├── 1-Projects/        → Active work (sprint, products, prospects)
├── 2-Areas/           → Ongoing ops (SOPs, agent onboarding)  
├── 3-Resources/       → Reference (Readwise, research, skills)
├── 4-Archive/         → Done/inactive

The twist? Symlinks. My agents were already referencing paths like ~/shared/sops/ and ~/shared/plan/. Instead of breaking those references, I moved the actual folders into PARA and left symlinks at the old paths.

~/shared/plan → 1-Projects/plan     (symlink)
~/shared/sops → 2-Areas/sops        (symlink)

Agents keep working. I get PARA. Zero breakage. This took 10 minutes.

The Semantic Search Layer

This is where it gets interesting. QMD indexes all your markdown files and lets you search by meaning, not just keywords.

Traditional search: "email security" → finds files containing those exact words. Semantic search: "what did we decide about protecting agents from phishing?" → finds the discussion about air-gapped email triage, even though it never uses the word "phishing."

I set it up on my Mac Mini (M4, 18GB RAM) and it indexed 98 files into 1,118 vector embeddings in under a minute. Entirely local. No OpenAI API calls. No data leaving my network.

A cron job re-indexes every night at 2am. New notes, new Kindle highlights, new SOPs — all searchable by morning.

The Readwise Integration

If you have a Readwise subscription, this is a 2-minute setup:

Install the "Readwise Official" community plugin in Obsidian
Connect your account
Set the base folder to 3-Resources/Readwise
Hit sync

Every Kindle highlight, every saved article, every tweet you've bookmarked — now sitting in your vault as searchable markdown. And because it's markdown in a shared folder, your AI agents can search it too.

I'm reading "Obviously Awesome" by April Dunford right now for positioning work. My marketing agent can now search my highlights when crafting outreach messages. That's not a workflow I designed — it's an emergent behavior of putting everything in the same folder.

The Real Win: Emergent Intelligence

Here's what I didn't expect. When your knowledge base, your task system, your SOPs, and your AI agents all share the same filesystem, things start connecting that you never explicitly connected.

My morning brief (auto-generated at 8:30am) pulls from:

Sprint priorities (plan/)
Overnight agent handoffs (overnight.md)
Infrastructure health checks
Token expiry warnings

It's not sophisticated AI. It's just... everything being in the same place, readable by the same tools.

The Second Brain isn't just for you anymore. It's for your entire team — human and AI.

How to Set This Up (30 Minutes)

Prerequisites: A Mac or Linux machine. Obsidian (free). Optionally: Readwise ($8/mo).

Step 1: Create the structure (2 min)

mkdir -p ~/shared/{0-Inbox,1-Projects,2-Areas,3-Resources,4-Archive}

Step 2: Open as Obsidian vault (1 min)

Open Obsidian → "Open folder as vault" → select ~/shared/

Step 3: Install plugins (5 min)

Settings → Community plugins → Browse:

Dataview — query your files like a database
Calendar — daily notes sidebar
Readwise Official — sync highlights (if you have Readwise)

Step 4: Multi-machine sync (10 min)

Install Syncthing on each machine. Point them at ~/shared/. They find each other automatically over your local network or Tailscale.

Step 5: Semantic search with QMD (10 min)

npm install -g @tobilu/qmd
qmd collection add ~/shared --name shared
qmd embed

Now you can run qmd vsearch "your question" from any terminal or AI agent.

Step 6: Connect your AI agents

Point your agents' working directories at ~/shared/ (or a subfolder). They can now read your SOPs, check your sprint, search your knowledge base, and write notes that you'll see in Obsidian.

The Bottom Line

Notion is a great tool for teams of humans collaborating in browsers.

Obsidian + filesystem is a better tool for solopreneurs collaborating with AI agents.

The files are yours. The search is local. The agents can read everything. And it costs nothing.

My Second Brain finally feels like it's actually part of my business — not just a place I visit to take notes.

I'm building a portfolio of 50 AI-powered micro-businesses by 2030. Follow along for the wins, fails, and everything in between.

Tools mentioned: Obsidian (free), QMD (free, open source), Syncthing (free, open source), Readwise ($8/mo), Tailscale (free tier), OpenClaw (AI agent platform)