Six schedules. One day. And the help inside it.

The problem underneath the problem

You already use AI for all of this. Just not in one place.

Ask anyone who leans on it. The gym routine is in one thread. The chapter they are stuck on is in another. What to cook, what to bring to the DMV, how to word the email — each one is its own tab, opened cold, explaining the same life from scratch every time.

None of those threads know it is Tuesday. None of them know the clinic runs until half twelve, that the kids get collected at four, or that the run has already been moved twice this week. So the advice is generically correct and specifically useless, and the person doing the reconciling is still you.

D.ai.ly is the other way round. The day comes first. The help comes from inside it.

Tuesday, and what each line opens into

Every block is a doorway.

Same six domains, same engine, same day. The difference is that you never leave to get the help, and the help never has to be told who you are.

What it looks like when two things want the same hour

One recommendation. And the reason.

Never a wall of options. One recommendation you can accept or decline, and at most two genuine alternatives — because 06:00, 06:15 and 06:30 is the same answer three times.

Why you can argue with it

The model writes. The engine decides.

This is the line the whole codebase is built to defend, and it is the reason D.ai.ly can show its working when nothing else in this category can.

• A proposal is reproducible

  Same day in, same recommendation out. Every time. A scheduler that answers differently on Tuesday than it did on Monday is one you cannot debug and cannot trust with anything that matters.

• A proposal is explainable

  "The clinic is immovable and the run has a daylight window" is a reason. "The model said so" is not. The reason is assembled from the constraint that actually decided it — and if a reason cannot be derived, the proposal is treated as wrong and never shown.

• Two immovable things produce no proposal at all

  Reported as unresolvable, rather than dressed up as advice the app cannot give. Saying nothing is sometimes the correct answer and almost nothing ships willing to do it.

• The retirement rule is arithmetic, not judgement

  Three declines of the same window closes it. That is counting, and counting does not need a language model.

So the model does the language — reading what you said in the interview, writing the brief, explaining the chapter, proposing the three dinners. It never chooses which block moves. That half is a deterministic engine with a test suite behind every rule on this page.

Two lives, same machinery

It asks who is answering first.

Nine questions before anything else. Three of them establish the person — a shift nurse and a self-employed designer do not get the same follow-ups — and the rest are generated from what is actually in your calendar, so it asks about your Tuesday shift rather than about a band you are not in.

• The doctor

  Clinic, the bills, the gym, the school run, dinner. The schedule part is not revolutionary and does not pretend to be — you already know the kids need collecting. What changes is the forty minutes after: three things you can actually cook in the time that is left, or the honest case for ordering something. The value is in the block, not in the block being drawn.

• The student

  Lectures, the gym, the chapter due Thursday. Say "six hours of study before Thursday" and the engine finds the hours — spread across three mornings, not crammed into one sitting nobody keeps — and tells you plainly when the hours are not there instead of inventing them. Then the study block itself is where the help with the chapter lives.

• The person starting something

  A business licence, a driving test, a first invoice. Each one is a date, a list of things to bring, and a sequence that has to happen in order. That is a schedule problem wearing a paperwork costume, which is exactly what this engine is for.

On screen

An editor's ledger, not a dashboard.

No cards anywhere. Hierarchy is carried by hairlines and exactly one heavy rule. Serif for anything written, mono for anything measured. Colour is reserved entirely for the six domains, so the interface itself never competes with your day for attention.

The boundaries

What it will not do.

• It does not sell your calendar

  There is no account to create. The reconciler runs on the phone. Reminders are local — no push server, nothing to breach, nothing stored.

• It does not gamify your week

  No streaks. No badge at 9pm for a habit you already know you missed. Two things earn an interruption: an unresolved conflict an hour out, and something immovable fifteen minutes out. Four a day, maximum, ever.

• It does not answer as if it knows things it does not

  Ask about something that is not on your day and it says so, then offers to put it there. That boundary is what makes the rest of the answers worth reading.

• It does not do your homework

  The study block helps you understand the chapter. There is a difference, and it is the whole difference.

Where it is

Early, and honest about it.

Both builds are real and running. Some of what you see is demonstration data until a calendar is connected, and where that is true the app says so on the screen rather than in the footnotes.

Android is on Play internal testing by invitation while the model layer is wired up. Ask and you will be added.