Now It Knows My Recipes by Name — and I Settled How I'll Talk to the Whole Thing

When I last wrote here, I’d just gotten the recipe finder working and ended on a wish: that one day I’d stop running these tools and start talking to them. This update is a few steps in that direction — and a decision about how to take the big one.

A quick reminder of what this is: a small system I built for myself to cook and eat a little better. I log meals by voice, find recipes from sites I trust, and plan a week against my nutrition goals. It’s not a product. It’s tools that fit my life, built in the open.

It knows my recipes now

Two changes made the day-to-day noticeably nicer.

First, I can ask any recipe in my library what’s actually in it. “What are the macros in my salmon bowl?” — and it adds up a single serving from real nutrition data and tells me. No more guessing whether a go-to dinner fits the day.

Second — and this is the one that makes me smile — I can just say the recipe’s name in a normal sentence. “I had my lemon-herb chicken and an apple for dinner.” It recognizes the chicken as one of my saved recipes, logs its real per-serving nutrition, looks up the apple on its own, and files both under dinner. One breath, done. I no longer have to remember a special command for my own meals; I just talk about my food the way I’d tell a friend.

A quiet weekly reckoning

I also gave it a weekly digest. At the end of a week it shows me how the week actually went — average calories and protein against my targets, where the food landed across the day — as a simple summary I keep for myself. It’s private by design; it isn’t posted anywhere. It’s the difference between hoping I ate well and knowing, without turning my life into a spreadsheet.

Teaching it to read (without stealing)

Here’s the part I’m proud of, because it’s where the ethics live.

Most cooking sites publish their recipes in a tidy, machine-readable form, and my finder has always read those cleanly. But some of my favorite cooks — Brian Lagerstrom is the one that finally pushed me — just write. Their recipes are free-flowing prose, with no structure for a machine to grab.

So I taught the finder to read prose too — but only the facts. It pulls out the ingredient list and nothing else. It never lifts the method, the story, or the writing, and it always credits and links back to the cook. The way I think about it: ingredients are facts; a cook’s voice is theirs. The tool is allowed to learn the first and forbidden from taking the second.

That line runs through the whole system, and it’s the thing I most want people to take from these posts: let the AI handle the language, keep the facts honest. Every nutrition number comes from the USDA’s database — the only source I’ll trust for that. The AI parses what I say; it is never allowed to invent a calorie. If it can’t find a real number, it flags it. It doesn’t make one up.

The big decision: how I’ll talk to all of it

Which brings me back to the wish I ended on last time.

I finally decided how I’m going to let myself talk to the whole system — not just the meal logger, but everything: finding recipes, logging what I ate, checking the week. In plain terms: a small helper that runs quietly on my own machine, which the AI assistant I already use can plug into. I get to converse with my kitchen — “what can I make with chicken and rice?”, “log what I just ate” — while every tool keeps running locally, on my data, with the same guardrails it has today.

I’m building it in careful steps rather than all at once, starting with some unglamorous cleanup under the hood. That’s on purpose. The fun part — actually talking to it — is worth getting right.

A few weekends in, this still isn’t a finished thing. It’s a system that gets a little smarter, and a little more honest, every time I sit down with it. That’s exactly what I hoped building in public would feel like.

I keep writing it down as it grows — the wins, the surprises, the dead ends. If that’s your thing, subscribe. Roughly weekly. No filler.