memorydial

A white onion almost broke me.

I was cooking dinner, logging as I went. I weighed the onion, chopped it, threw it into the pan, then opened MyFitnessPal. The first result had more fat than a stick of butter.

I knew that was wrong. I spent the next five minutes hunting through a database of user-submitted shite, guessing which entry was accurate. For an onion.

Why am I still using this crap?

Right then I decided to build my own.

It came together with Claude Code pretty quick. A Django Web app, SQLite Database and a cheap GPT model deployed onto a raspberry pi and made accessable over the web via Tailscale.

You describe food in plain English. It returns reasonable nutrition data. Just type what you ate.

When I'm cooking, I know what's going in. Last week I made chili. I pasted the ingredients straight from a YouTube description: ground beef, kidney beans, tomatoes, onion, half a packet of American cheese. Told it six portions. It came back: 340 calories, 26g protein per bowl.

When I'm out, I don't know what's in the burrito. I don't need to. I type “chipotle chicken burrito, no sour cream” and it comes back: 650 calories, 42g protein, 80% confidence. Good enough. Logged and done.

Here's what the big apps miss: a rough estimate you record beats a precise measurement you don't. Tracking is a habit. Habits need to be easy.

Last week I wanted to re-log yesterday's breakfast with one click. Took an hour. Now it's there, because it's mine.

And mine doesn't store passwords in plain text.