What It's Like to Feed Yourself to an AI
AI · Productivity
Last weekend I asked Claude to help plan my week. Gave it zero context. It came back with: "You worked 45 hours last week, your rest ratio is dropping, and since you're flying to Korea soon, I'd suggest clearing your high-priority tasks first."
That feeling of being seen through a screen. Honestly? Kind of great.
It Started Seven Years Ago
Seven years ago I started logging every hour of my day. No grand plan — I just wanted to gamify my life a little. Seeing where my time went gave me this small hit of satisfaction, like scoring my own day. Back then there was no AI in the picture. I was summarizing everything with Excel formulas, building dashboards by hand.
Then AI happened.
Markdown Is an LLM's First Language
I realized that if I synced this data into my note vault as markdown files, AI could read it natively. No database, no API integration needed. A single line like 09:00-11:30 | W and it immediately understands you worked for two and a half hours that morning. Markdown is an LLM's first language.
So I set up an automated pipeline. Every day, my time logs, financial data, chat history, and task lists from Dida365 all sync into the vault automatically. A quiet batch of files, updating themselves daily, waiting for AI to read them. I also do light maintenance — dropping in scattered thoughts, research notes, things I've read.
As the data grew, the AI started knowing me better. A couple days ago it noticed that some tasks on my list hadn't moved in weeks, and that I was spending consecutive hours every evening doing things unrelated to my plans. It told me straight: "You're interest-driven. You need to design around that, not against it."
Uncomfortably accurate.
One Data Center, Many Frontends
Looking back, the architecture is simple: one data center, many frontends. The data center is the vault — everything flows here. The frontends are tools I vibe-coded myself: a time tracker, a budgeting app, a task manager. Each one built exactly the way I want it. Want gamification? Build it. Want a dashboard that shows today at a glance? Build it. No compromising with someone else's product logic.
Every AI conversation starts from this vault. It can see my time records, finances, task lists, work conversations, reading notes, and journal entries.
The Feeling of Being Remembered
There are these moments in conversation where it pulls up something you wrote months ago, or connects an observation from three months back to what you're dealing with now. And you think — oh, it actually remembers. That feeling of being remembered is honestly the thing I like most about this system. Not the productivity gains. Not the dashboards. Just having something that continuously knows you.
The biggest side effect so far: sometimes I wonder whether understanding a person actually requires consciousness.
hi-time and hi-money are the frontends I built for this system. More details on the project page.