Why Repeat Eat works with your AI

The first time you use AI with Repeat Eat, the process may seem unusual. The app prepares a photo and a detailed instruction, you send them to the AI service you prefer, and then you bring the structured result back to the journal.

It takes a few more taps than a hidden AI button. That boundary is deliberate.

A journal and an AI have different jobs

A food journal should remember exactly what you chose to save. An AI looks at incomplete evidence and makes an estimate. From a photo it may recognize rice and chicken, but it cannot reliably know the weight, the amount of oil, or what is hidden underneath.

Repeat Eat keeps those responsibilities separate. The AI proposes a useful starting point. The app keeps the record. You decide whether the proposal is close enough, needs a correction, or should be ignored.

Why the AI is not built into Repeat Eat

I did not want the journal to depend on one model or one company. People have different preferences, existing accounts, supported services, and privacy requirements. AI services also change much faster than a personal journal should.

Using the service you choose avoids tying your food history to a particular provider. It also means Repeat Eat does not need an account, a developer-run AI server, or a recurring fee for every meal you analyse.

Most importantly, sharing is an explicit action. Your photo remains in the local journal unless you choose to send or copy it to an external AI. That service then handles the material under its own terms and privacy policy. Repeat Eat does not silently upload the photo for analysis.

BYOAI: Bring Your Own AI

I think of this approach as BYOAI, or Bring Your Own AI. Instead of making one model part of the product, Repeat Eat lets you use the AI you already know and trust. The app acts as a bridge between that assistant and your journal.

This is different from BYOK, or Bring Your Own Key. You do not need to find an API key, paste it into Repeat Eat, or give the app direct access to an AI account. You simply use the normal app or website of the service you choose.

Why I chose an open format

I arrived at this design after trying to use an AI chat as the food diary itself. The chat was useful for describing a meal, but unreliable as long-term storage. It could confuse days, forget earlier entries, and make the history difficult to inspect or correct.

The useful part was the estimate, not the chat around it. I needed a clean handoff from a temporary conversation to a journal that could keep the result.

A conversational answer is pleasant to read, but difficult for a journal to use safely. Repeat Eat asks for a small JSON object with specific fields for the meal, serving, protein, fibre, and calories.

The instruction and response use ordinary text and JSON rather than a private protocol. A capable AI can produce the format, a person can read it, and the result can be copied or corrected without depending on a hidden Repeat Eat service. The same principle applies to journal exports: your records remain available as a portable file instead of being trapped in an account.

Open does not mean unchecked. The format has a version and stable field names, so Repeat Eat can verify the structure and reject data it does not understand. It cannot prove that an estimate is nutritionally correct, but it can avoid quietly treating an arbitrary paragraph as a valid journal entry.

Why you review the result

A confident answer is still only an estimate. Before saving, Repeat Eat shows the dishes, portions, and totals returned by the AI. If something is wrong, you can tell the AI that the portion was smaller, use calories from a visible label, or correct the data before it becomes part of the journal.

I prefer this to pretending that food recognition is automatic or precise. The aim is not to remove judgment. It is to reduce the work of describing an unfamiliar meal while keeping judgment in the right place.

The complete flow

1. Take or choose a meal photo.
2. Let Repeat Eat prepare the photo and a precise instruction.
3. Share them with the AI service you choose.
4. Check the suggestion and ask for corrections if needed.
5. Import the structured response, review it, and save it to your journal or library.

Once the meal is familiar, you may not need AI again. Save it as a reusable set, repeat it next time, and adjust only what changed. That is the larger idea behind Repeat Eat: use AI when it removes uncertainty, then let memory and repetition make everyday logging fast.