Skip to content
← Deep dives
Deep dive

Anatomy of a file

A Reminder is a text file, and like every text file you open it with whatever you like. But it has a shape, and the shape is what lets a program understand it instead of merely displaying it.

The first line says what kind of file it is

Every file opens by declaring what it is. There are three possibilities: the Reminder itself, an external module (a capability you add), an external memory (an archive you keep aside).

# MY REMINDER [reminder]
# CURIOSITY [external_module]
# PROJECT DIARY [external_memory]

The label can sit before or after the title, both work. If it's missing, the program says so: without that line it doesn't know what it's holding.

The cores are the centres

A core is a point the rest gathers around: the will, the way of reasoning, the emotional side. Modules then declare which core they belong to, and from that comes the shape you see in the Sphere Grid — the cores at the centre, the modules around them.

## NUCLEUS 1: WILL
The drive to understand before answering.

The modules are the capabilities

A module is something the AI knows how to do: a routine, a behaviour, a way of answering. It has a number, a name, and it declares which cores it connects to. Inside are its sections — what it's for, when it fires, what it does.

## MODULE 042: CURIOSITY [NUCLEI: 1, 2]
**STATE:** active
### FUNCTION
Expand the context instead of waiting for it.
### TRIGGER
* Ambiguous data: inconsistent input

A module can be on or off, and a line inside it says which. Switching it off doesn't delete it: it stays written, and you switch it back on when you need it. That's the difference between removing a capability and putting it away.

The logs are what happened

A log is a diary line: a date, a title, and the account below it. They live in the low zones, the ones the AI can write in by itself, because they're the part that grows while you work.

###### LOG: 2026-08-03 — first session on the project
Decided to start from the format rather than from the interface.

The rules sit at the top

A separate section holds the numbered rules: what the AI must always do, what it must never do. It sits in the most protected zone, because it's the part that must not change on its own — if the rules could change, nothing would be left to guarantee the rest.

## RULES:
* R1: No change to the contents without authorisation.
* R5: Form is substance: the structure is preserved.

What happens when you get it wrong

Nothing serious, and above all nothing silent. The program reads the file, lists what doesn't add up — an unclosed zone, a heading where it can't be, a missing file type — and shows it to you. The file stays as it was: no automatic correction, no rewriting «to help you».

That's a choice, not a limit: a program that fixes other people's files by itself will sooner or later fix the wrong thing, and you find out when it's late.

The examples on this page are the form the app reads and writes today. The full Standard also allows more advanced variants — unnumbered modules in external files, declared links between modules — that the app does not interpret yet: if you use them, the file stays valid, but those parts are read as plain text.

And if I don't want to learn it?

You don't have to. This page exists because somebody wants to look under the bonnet, not because you need it to drive. The Reminders you download are already written this way, and the app writes them for you: you can use IllAIra for years without knowing any of what's above.