My LifeOS Fits Into Seven Domains

LifeOS is now small enough to draw.

At the top level, the whole system fits into seven domains. These are the areas of life where context needs to stay stable enough for an assistant to work with it.

System Global rules, source hierarchy, memory boundaries, routing, and assistant behavior.

Health Labs, checkups, supplements, training, recovery, and health signals.

Nutrition Food choices, logs, recipes, restaurant decisions, and delivery tradeoffs.

Lifestyle Home, routines, relationships, mood, environment, and quality of life.

Work Projects, evidence, decisions, communication, and professional history.

Public Work Public notes, positioning, offers, and work that can be shown outside the private system.

Finance Money, assets, property, taxes, budgets, documents, and financial decisions.

For another person, this map might have eight or nine domains. Maybe ten. I suspect ten is close to the useful upper limit. Above that, the structure starts demanding more attention than the work inside it.

Local rules

These domains are more than labels.

Each one has its own folder, source files, context, routing rules, working principles, and local AGENTS.md. That local file tells the assistant how to treat the domain: what belongs there, which sources have authority, and what kind of update may be needed.

A health question and a work communication problem require different context. A finance document should stay connected to financial evidence. A public positioning idea should remain separate from private work details. A temporary mood or travel state should not quietly become permanent memory.

Domain routing gives the assistant a useful first move: know where to look.

From there, it can identify what it is dealing with: raw evidence, a durable fact, a current open loop, a temporary condition, a draft, or a decision that should update the source of truth.

This is where LifeOS starts to feel like a working context system rather than a collection of notes with AI attached.

Why build it this way

ChatGPT is currently the interface I use most often. The system itself lives outside it.

Its working state is stored in readable files: context, evidence, decisions, local rules, and update history. I can inspect what the assistant was given, correct it, and carry the same structure into another model, an API, or eventually a local LLM.

Built-in AI memory is difficult to audit. You rarely know exactly what was saved, what was compressed, or which old detail is shaping an answer. LifeOS makes that layer explicit. The model still does the reasoning, but it reasons against context that has an owner, a source, and a place to be corrected.

This does not guarantee a good answer. It makes good answers more repeatable, and bad ones easier to diagnose.

The AI can change. The operating state remains mine.