Why I Built Harbor: The Case for a Private Knowledge Base
Every few months, I'd lose something I couldn't afford to lose.
Not data — the files were always there. I'd lose the context. A conversation where a colleague mentioned a preference. A note I wrote after a hard meeting. A decision I made for a reason I couldn't reconstruct six months later. The information existed, scattered across Notion, Apple Notes, a few text files, and some half-remembered Slack threads. But none of it was findable in the moment I needed it.
That problem sounds small. It's not.
The real cost of scattered knowledge
When context is hard to find, you stop trying to find it. You start from scratch more often than you should. You make the same decisions twice. You forget what you promised someone or what they told you.
For individuals — people who work with AI tools, manage projects, think carefully about their craft — this is a compounding problem. The more ambitious the work, the more expensive the forgotten context.
I tried to fix it with existing tools. Notion is capable, but it becomes a graveyard of pages you stop maintaining. Obsidian is better for personal notes, but it doesn't structure information in a way AI can reliably query. And AI tools themselves — Claude, ChatGPT — have no persistent memory of who you are, what you care about, or what you decided last week, unless you tell them every session.
What I actually wanted
The thing I wanted was simple to describe and hard to build: one place where I write things down, and everything I write stays organized, searchable, and usable — by me and by the AI tools I trust.
Not a second brain. Not a productivity system. Just a calm, private workspace where knowledge compounds instead of decays.
I wanted it to be:
- Readable without the app. Plain Markdown, not a proprietary format. If Harbor disappears tomorrow, I still have my files.
- Structured enough for AI. People, tasks, projects, preferences — stored in a format that AI tools can query without hallucinating.
- Auditable. Every time an AI changes something, I see the diff. I approve or reject. No silent edits.
- Self-hostable. My knowledge should live on my infrastructure, not someone else's.
What we built
Harbor is the result of working backward from those requirements.
The core idea is a Markdown-first knowledge base with a structured layer underneath. Documents are stored as .md files on disk — readable in any text editor, exportable as a .zip anytime. But structured data like people, tasks, and preferences are also represented as typed blocks within those documents, backed by a SQLite database that stays in sync.
The AI integration works through MCP — the open protocol that lets AI tools read and write to systems they're connected to. When Claude or ChatGPT needs to look something up about a person I know, it queries Harbor's MCP server. When it wants to add a task, it proposes a patch. I review it. I approve it. Nothing is written without my say.
This sounds like a lot of ceremony. In practice, it takes two seconds and means I always know what changed.
What's next
Harbor is in private beta right now. We're working with a small group of early users — mostly people who think carefully about knowledge, context, and how AI fits into their working life.
If that sounds like you, get early access or self-host it for free.
I'll be writing here about what we're building, what we're learning, and the harder questions underneath it all — about how people should store knowledge, who should own it, and what it means for AI to have access to it.
Asgeir Albretsen is the founder of Harbor.