← built & broken
2026-06-04 · 3 min read

My Personal OS

I'm a forgetful person. Never journaled, never took notes — just tried to keep everything in my head. That doesn't work. So I built a system.

personal-osbuildingai

I forget things. Not the dramatic kind — not names or faces — but the quiet kind. Months go by and you realise you never followed through on something you cared about. You keep making the same mistake. You are aware of who you are but somehow cannot change the things you don't like.

Before this, I tried nothing. That is the honest answer. No Notion, no journaling, no habit trackers. Just my head. I figured if something mattered, I would remember it.

That doesn't work.


What I built

In late 2025 I started building a personal operating system. Not productivity software — I have seen those, and they require you to maintain them. This is different. It is an Obsidian vault where an AI (Claude, by Anthropic) maintains the knowledge layer. I feed it raw material. It writes the wiki.

The vault has three layers:

  • Raw sources — immutable input. Daily journal transcripts, work documents, exported data.
  • Wiki — Claude-maintained. Entities (people, projects, habits), concepts (frameworks I keep returning to), processed daily notes.
  • Working layer — in-progress thinking. Drafts, extracts, scratchpad notes before they earn a wiki page.

Every evening a Telegram bot asks me about my day. Spending, mood, what I worked on. I answer in short messages, the way I actually talk. The bot processes the session, Claude synthesises it, and by morning there is a new daily note in the wiki. Transactions logged. Patterns flagged. Open loops tracked.

I do not write any of it. I just live and answer.


What it has actually done

Here is the honest version: I have always been aware of who I am. That was never the problem. The problem was change. Knowing you procrastinate is different from not procrastinating. Knowing you spend badly is different from stopping.

What the OS does is make the pattern visible, consistently, without me having to remember to look. Open loops do not vanish when the conversation ends. Habits I said I would start are still there three weeks later, quietly flagged. The mirror does not blink.

Whether I act on it is still on me. But at least I cannot pretend I did not see it.


The Ecosystem Hub

The more systems I built — journal bot, finance dashboard, monitoring stack, portfolio site, data pipelines at work — the more they started to drift from each other. Something changes in one place and three others are quietly wrong. You only notice weeks later.

The Ecosystem Hub is my answer to that. It is a reflex, not a tool. You do not call it. Every time I end a Claude Code session, a hook runs a reconciler that reads what changed, updates a state file for each tracked project, and stages proposals if one change implies action somewhere else. Next session, it surfaces a digest: "saw X, updated Y, two proposals pending."

The hub is the connective tissue. Not just inside the personal OS — across every project I am running. Side projects, deployed apps, work tools. One source of truth, maintained automatically as a side effect of working normally.

It went live two days ago. 174 tests, seven sessions of TDD. The fact that I built it with that level of rigour for a system designed to help me stop forgetting things is not lost on me.


The ongoing question

Building this taught me something I did not expect: the system surfaces things I already knew. The insight was not the data, it was the consistency. When the same thing shows up week after week in a system that never forgets, something shifts.

Not fixed. Just harder to ignore.