How I keep track of 153 AI projects

In short

I have 153 projects, 87 of them live online. Three things keep me from getting lost. One place where every project has a card, and AI fills it in. A clear call on what lives under one roof and what stays on my own machine. And a Google login in front of every private project.

Worked: cards written by AI. Did not work: filling them in by hand.

When I tell people how much I build with AI, the question I hear most is not "how" but "how do you not get lost?". I have asked other people the same thing more than once. The most common answer was honest: I don't. Projects pile up faster than you can remember them.

Today my list holds 153 projects. 87 are live online, 66 run only on my computer. 37 of them are experiments built in a day or two, and some of those I may never open again. Without a map I would have lost half of them already.

Date: 09/10/2026
Tools: AI Dashboard (my own app on my computer), Claude Code, GitHub, Cloudflare Access
What was checked: Project counts taken from the list itself on the same day
Not for: If you have three projects, plain notes are enough

One place where I see everything

AI Dashboard is a simple app on my computer. Categories on the left, the project list in the middle, the selected project's card on the right: what it does, why I built it, whether it is live or only on my machine, a few screenshots and a log of what shipped when.

AI Dashboard window: categories on the left, a project list with icons in the middle, counts at the top: 28 projects, 18 local, 10 live. — The project list. For this screenshot I kept 28 of the 153: no clients, no family, no health. AI Dashboard, 09/10/2026.

The important part is not the app. It is that I don't fill it in. I used to paste the same prompt into every project and carry the answer over to the card by hand. I did that maybe eight times and knew I would not keep it up. Now, when I finish a session in Claude Code, it reads the project's code and writes or updates the card itself. New feature, moved online, archived: the card catches up. Each project folder also keeps a changelog tied to the card, so what I write in one shows up in the other.

AI Dashboard project card: name, Open, Copy info and Edit buttons, three screenshots and the sections Functionality, What it does, Why I use it. — The card for AI Dashboard itself. Claude wrote it after reading the code. 09/10/2026.

Under one roof, or only on my machine

The second thing took me a while: not every project should live in the same place.

Some are best with everything under one roof. Code in a GitHub repository, the site, the database and the domain in one Cloudflare account. I get history and a backup, and a change goes live with one command. Public sites, lab experiments and training pages live like this.

Others are best kept on my computer only. Anything holding health numbers, finances or family details never goes to GitHub, not even to a private repository. Those projects have local version history and nothing else. If the laptop disappeared I would miss them, but they are not sitting on anyone else's server either.

The dashboard is the one place that knows about both groups. The projects live in different places, the list is one. That is the roof I actually needed: not one place for every project, but one place that knows where each project is.

A private project with a normal link

The third piece came out of a mistake. I have a page that checks real flight prices every 10 minutes for trips I am watching. Its first version ran on a public address, so anyone with the link could see the travel plan. It also showed made-up prices and cost more than any other project I had on Railway.

Now Cloudflare Access stands in front of it. You open the link and the first thing you see is a Google login. One account gets in, mine. I did not build a login page, I store no passwords, and it is free for up to 50 users.

AI Dashboard card for Flight Deal Scanner: what the page does, why it is private now, tags cloudflare-workers, d1, cloudflare-access, travel and private-tool. — The card also says why the project is private. 09/10/2026.

Google sign-in window reading: Sign in, to continue to cloudflareaccess.com. — What anyone else sees when they open the same link. 09/10/2026.

That changed how I build. "Can this be public?" used to stall me. Now private is one setting, not a separate job.

It is not a Claude trick

I use Claude Code, but nothing here depends on it. The card step is one written instruction in the project: when the work ends, read the code and update this entry. ChatGPT Codex reads the same kind of instruction file (AGENTS.md), and Cursor and Gemini CLI have their own. Versions on GitHub, a folder that stays on your computer and a Google login in front of a private page work the same whichever tool wrote the code. The dashboard itself can be a plain spreadsheet or a text file. Mine is an app because I built one, not because you need one.

What does not work

AI Dashboard lives on one computer, so I cannot see it on my phone. The window does not refresh itself: when Claude writes a card, I have to reload it. The app does not catch duplicates, so the AI has to check whether a project is already on the list, or a second card appears. And I only refresh screenshots after bigger changes, so some of them show yesterday's version.

The fair objection: maybe the problem is not how to keep track of 153 projects but that there are 153 of them. Partly true. A list does not make projects go away. But it shows me which 37 are experiments I can safely forget, and which 87 are live and need to stay in order.

What to do on Monday: put your projects in a plain table with four columns: what it is, what it does, where it lives (my computer, GitHub, online) and who can open it. Then ask your AI to update it at the end of every work session.

The rule

A project you can no longer find barely exists. Building is cheap now. The hard part is finding what you built three weeks later, and remembering who you let see it.

What the websites I build this way actually cost is written up separately. Small projects that can be public go to the lab.