How we keep our gym and meal log with AI

On the Traiker team we keep our gym and meal log by talking to an AI assistant connected to our own database: we tell it what we have trained or eaten in plain language, it saves it in order, and later we ask it about progress or what to have for dinner. No conventional fitness apps or endless forms.

We're telling you how we do it, because it is the use case that gave rise to Traiker and the one we use most every week.

Why did we leave the usual apps behind?

Like almost everyone, we tried several. The problem is usually the same: you have to adapt to their way of logging, and food lives in a different app from training, so crossing the two ends up in spreadsheets.

What we wanted was simple: to log quickly, in our own words, and to be able to ask questions that mixed training and food.

How do we log a workout?

When we finish, we write something like “leg day today: squat and deadlift, four sets each” with the approximate weights. The assistant interprets it and saves it as structured records: exercise, sets, reps, weight and date.

This works because the assistant doesn't save anything “in its memory”: it calls a tool in our database through MCP. If we use another compatible assistant tomorrow, the records are still there.

And meals?

The same. Something like “meal: pasta with chicken and salad” is enough. There's no need to weigh every gram; a consistent, approximate log is worth more than a perfect one you abandon after two weeks. If a recipe repeats, it is saved once and then you just say its name. What does help is logging at the time, or at most that same evening: by the next day the details are forgotten and the log loses value. Consistency matters more than precision.

What do we ask it afterwards?

Here's the fun part. Because training and food live in the same place, we can ask things like:

“How has my squat changed this month?”, “what should I have for dinner if I have a hard workout tomorrow?” or “on the days I perform worse, had I eaten too little?”. The answer comes from our own records, not from generic advice on the internet.

The exact numbers don't matter, because every body is different. What matters is that looking at your own data with simple questions makes it easier to stay consistent than any pretty chart, and it is harder to fool yourself about how much you are really training.

What have we learned?

That logging has to cost less than not logging. That structure matters: if the assistant saves “bench press” one day and “bench” another, the questions fail, so it's worth having a clear exercise schema. And that it's worth reviewing what has been saved from time to time, because the assistant sometimes misreads a weight or a set. And that you don't need to log everything from day one: starting with workouts only and adding meals once the habit is established works much better than trying everything at once.

That schema of exercises, meals and recipes is exactly what we want to offer as a template in Traiker, so nobody has to build it from scratch.

Frequently asked questions

Do I need a fitness app for this?

No. All you need is an assistant compatible with MCP and a place to store your data with a clear structure.

Does the AI build my routine for me?

It can suggest things based on your records, but it does not replace a coach or a health professional.

What happens if the assistant logs something wrong?

It happens. That's why it's worth reviewing the records from time to time and making tools that write data ask for confirmation.

If you want to try it with your own workouts, sign up for the beta.