A kill switch for every AI automation

At some point I couldn't answer a simple question: what is allowed to run right now, and what can it touch?

Too many automations to track

When you build things quickly, they pile up. Each one made sense when I added it. Together, they were more than I could hold in my head.

I wanted a switch for each one.

A Mac app with a switch for each automation

I made a Mac app of my own. Every automation and every tool connection had its own on/off switch, and I could sort them by company, by client, and by what's personal.

The part I cared about most: turning something off actually stopped it. Flip an automation off and it doesn't run on its next turn. Flip a connection off and my AI tools get blocked when they try to use it.

Keeping the app alive

The app worked, but a custom app is one more thing to keep alive. It had its own data format, its own bugs, and its own updates. When I changed how things were organized, the app had to change too. I'd built a tool to reduce mess and given myself something new to maintain.

Files as the source of truth

The switches and the facts already lived in plain files. The app was just a screen on top of them.

So the files stay the source of truth, and the blocking still lives there, so a switch still really stops things. The screen on top can be anything, including a page in a tool I already use. The custom app is on its way out.

Notes for next time

• Make the switch realA switch that blocks the thing is worth more than a status light.
• Don't build a tool you have to babysitIf a tool you already use can be the screen, use it.
• Keep the truth in one placeScreens come and go. The files stay.