Most Personal Knowledge Management (PKM) systems fail for the exact same reason home gyms collect dust: the maintenance overhead exceeds the daily value.
You spend three weeks down an internet rabbit hole designing the "perfect" note-taking architecture. You color-code tags, configure Dataview tables, tweak theme CSS, and marvel at your dazzling, celestial graph view. Then real life happens. Tax receipts pile up on the counter, three client deadlines hit at once, the tractor needs a hydraulic hose replaced, and opening your note app suddenly feels like clocking in for a second, unpaid shift as a digital museum curator.
So you stop opening it. The notes freeze in time. The junk drawer wins.
For years, I struggled with this exact gravity. I think in systems, data flows, and code. My wife and creative partner, Ivy, thinks in story, human resonance, and narrative connection. Together, we operate Camwyn & Co. as an independent house of projects—incubating software like ManyHats Ledger, stewarding heritage livestock at Mema's Inclination, and running literary advisory work at Poetic Entanglement.
We desperately needed a shared, single source of truth across our ventures. But I knew that if that system required Ivy (or anyone else on our team) to manually wrangle Markdown formatting, manage Git merge conflicts, or memorize folder taxonomies, it would die on the vine within a month.
We didn't need another app to look at. We needed a headless brain.
Doing this setup yourself, with no one else to share it with? A few sections below assume two people. If you're the one who'll actually run the setup steps (Step 2's script, Step 3's config editing), you can skip "From Conversational Writers to Autonomous Coding Agents" (that's about keeping two people's parallel workflows in sync) and skip "One More Board on the Bench" and Part Two entirely (the hosted, multi-user build only solves a sharing problem you don't have). On one device? Skip the sync step too (Step 1 under "The Realization").
"The Real Problem: Don't Double the Noise" is still worth reading solo — it's not about sharing, it's about what your own mind needs from a tool. And if you're the non-technical reader hoping to skip the terminal entirely: I'll be honest that Steps 2 and 3 below still require running a script and editing a config file. That's the real bar, not whether you're sharing a vault. If that's not something you want to do yourself, you're exactly the case "The Real Problem" describes — the fix isn't a simpler guide, it's having someone technical set this up for you once. Everything after that, in "How to Use It Daily," works the same for you as it does for Ivy.
The Discovery: Courtney, Tiago, and Two Very Different Brains
The conceptual breakthrough didn't happen in a vacuum. It started with my friend Courtney Robertson, who wrote insightfully about cutting through brain fog by using LLMs as an autonomous synthesis layer over an Obsidian wiki.
Courtney's article introduced me to Tiago Forte's P.A.R.A. Method (Building a Second Brain). The moment I understood PARA, I fell in love with it. Tiago gave the world a brilliant, pragmatic taxonomy for organizing digital life by actionability rather than academic subjects:
- Projects: Active initiatives with fixed finish lines.
- Areas: Long-term domains of responsibility with continuous standards (brand identity, farm operations, finances).
- Resources: Reference libraries and playbooks.
- Archives: Cold storage for completed milestones.
For my systems-architect brain, PARA was poetry. It made complete structural sense.
The Real Problem: Don't Double the Noise
Then came the real-world collision with human nature.
Ivy is non-technical, but more importantly, her brain is already a living, intuitive mind-map. She naturally holds narrative arcs, emotional resonance, and brand connections in her head without needing a screen to tell her where they connect.
When she looks at software like Obsidian—particularly the sprawling visual graph views, complex plugins, and nested folder trees—it doesn't bring clarity. It just doubles the noise. To her, it felt like putting an unnecessary layer of software bureaucracy over a creative process that already worked effortlessly in her mind.
She didn't want to learn Obsidian. In fact, she didn't want to adopt any new software. Her own internal mental model worked just fine, thank you very much.
And she was completely right.
We wanted her to have full access to the collective "brain" of our studio—to query our brand guidelines, check project statuses, and record dispatches. But forcing her to adopt an app that conflicted with how her mind naturally operates would have been a catastrophic failure of system design.
So we flipped the question on its head:
What if we built a Second Brain so seamless that she never had to open Obsidian at all?
The Realization: The Filesystem Is the Database
Obsidian's greatest architectural genius is not its desktop interface. It's the fact that an Obsidian vault is just a clean, local directory of plain Markdown (.md) files sitting on your physical hard drive.
No proprietary database binary. No cloud lock-in. No SaaS subscription holding your life's work hostage.
Because the vault is just a folder of text files, we realized we could bypass the user interface entirely using the Model Context Protocol (MCP) and Steven Stavrakis's brilliant obsidian-mcp server.
+-------------------------------------------------------------+
| Ivy's Computer |
| |
| +---------------------+ +--------------------+ |
| | ChatGPT Desktop | <------> | obsidian-mcp | |
| | (What Ivy sees) | MCP | (Headless Server) | |
| +---------------------+ +---------+----------+ |
| | Reads/Writes |
| v |
| +--------------------+ |
| | Synced Vault Files | |
| | (No App Running!) | |
| +--------------------+ |
+-------------------------------------------------------------+
Here is what happens in practice:
- Invisible Sync: We keep our master vault synchronized across machines in the background using End-to-End Encrypted (E2EE) Obsidian Sync.
- Headless MCP Connection: We hooked
obsidian-mcpinto our AI desktop clients (Ivy uses ChatGPT Desktop on Windows; I use Claude Desktop and Antigravity IDE). - The Zero-Friction Experience: Obsidian does not need to be open on Ivy's computer. She simply opens ChatGPT, which she already uses every day, and talks to it naturally:
"Can you search our Camwyn vault and give me the brand colors and approved taglines for ManyHats?"
(ChatGPT silently queries the vault via MCP tools and responds with the exact hex codes).
"Draft a 2-paragraph dispatch about winter preparations for the orchard and save it to Dispatches."
(ChatGPT reads our brand voice rules, writes the essay, and callsobsidian_create_noteto create the.mdfile on her disk).
Two seconds later, that new dispatch syncs to my machine. Neither of us had to touch a file tree or fiddle with YAML.
Prerequisite: Connecting custom MCP servers in ChatGPT Desktop requires ChatGPT Plus, Pro, Business, Enterprise, or Edu — it's not available on the Free plan. (Separately: ChatGPT's own upload-based Skills feature is Business/Enterprise/Edu-only, but this setup doesn't use it — the
AI CONTEXT.mdgrounding here comes from a pasted custom-instructions block, available on any paid plan.)
The Secret Sauce: The AI Constitution (AI CONTEXT.md)
Why doesn't the AI generate generic, corporate-slop filler?
Because at the root of the vault sits a single file: AI CONTEXT.md.
This note acts as the master constitution for every AI worker that touches our knowledge base. It outlines:
- Our Operating Thesis: Smart can be warm. Ambition can have soul. Serious work can still be fun.
- Ubiquitous Vocabulary: Exactly what words we use—and which marketing buzzwords ("synergize", "leverage", "game-changer") are strictly banned.
- Editorial Archetypes: Guidance on when to speak in the Craftsman voice (tactile, build quality, workshop joinery) versus the Storykeeper voice (hospitality, memory, human connection).
- Operational Guardrails: Rules requiring the AI to read before editing, preserve frontmatter schemas, and append timestamped worklogs with author attribution (
### 2026-09-08 — @ivy).
When an AI worker is anchored to an authoritative constitution and has hands into your project history, it stops acting like an erratic chatbot and starts acting like an aligned team apprentice.
From Conversational Writers to Autonomous Coding Agents
What makes this headless architecture so powerful is that it doesn't just work for conversational desktop chats—it scales all the way up to autonomous engineering.
On my side of the workbench, our IDE agents (running in Google Antigravity, Claude Code, and Cursor) connect to the exact same vault using custom agent skills. While Ivy is using ChatGPT to brainstorm dispatches or review brand voice, my coding agents are autonomously reading Architectural Decision Records (ADRs), keeping sprint worklogs updated, and re-indexing master Maps of Content (MOCs) directly from our active Git codebases.
Crucially, because the coding agents read AI CONTEXT.md and our brand directives, they don't just write functional code in a vacuum—they actively enforce our design tokens, respect our architectural constraints (like strict 0px border radius), and craft UI microcopy, button labels, and error states that speak in the exact same grounded, hospitable brand voice that Ivy curates.
Two completely different people, working in two completely different environments, interacting with the exact same shared memory—with zero manual data entry or software friction on either side.
Building the Jig (And Giving It Away)
In a woodshop, when an off-the-shelf tool doesn't make the cut, you don't buy a factory. You make a jig.
Once we got this headless brain running smoothly between our own desks, we realized how many solo builders, creative studios, and multi-disciplinary teams are suffocating under the weight of disorganized notes and rigid productivity apps.
So we packaged the entire architecture into an open-source starter kit:
👉 Headless AI Second Brain Starter Kit on GitHub
The repository includes:
- The Sanitized PARA Skeleton: Pre-structured folders (
00-INBOX/,Projects/,Areas/,Resources/,Archives/) with starter MOCs and Dataview task queries. - The AI Constitution Template: A fill-in-the-blanks
AI CONTEXT.mdto establish your brand voice and domain definitions. - Plug-and-Play Connectors: Exact form setup instructions and JSON configuration files for ChatGPT Desktop, Claude Desktop, and Antigravity / Cursor.
- 1-Click Setup Scripts: Automated PowerShell and Bash installers for Windows and macOS.
You can click "Use this template" on GitHub, run the 10-second setup script, paste your vault path into your AI assistant, and start having real conversations with your knowledge base in under five minutes.
One More Board on the Bench
Everything above is Phase 1 — the version any solo builder or small team can stand up in an afternoon, with a synced vault and a paid Obsidian Sync subscription, no server to rent and no infrastructure to babysit. It's still exactly what we'd tell someone to start with.
Since writing this, our own setup outgrew it. Two people, two vaults, and the question of what happens when someone's laptop is closed pushed us toward something bigger: a shared, hosted vault that Ivy and I both reach the same way, from anywhere, with no local server tied to either of our machines. Building that — and everything that broke on the way there — is Part Two.
Tools That Get Out of the Way
Software should never demand that you become someone else just to use it.
If you are a systems architect who loves terminals, you should be able to live in the terminal. If you are a writer or strategist who loves natural conversation, you should be able to chat without opening a code editor.
A good tool doesn't force everyone at the table to hold the hammer the same way. It just holds the workpiece steady, makes the cut clean, and gets out of the way so you can get back to the work that matters.
