About WeatherSpirit

Why this exists, and who is making it.

WeatherSpirit began with an argument about a blanket.

The problem was never the forecast

Anyone who keeps a horse has had the evening conversation. It is going to be 38 tonight, which sounds cold. But she is clipped, and she is going out, and it was 60 today so she will still be warm, and the 200 might be too much — and by morning, if you got it wrong in the generous direction, she has been sweating under a rug for six hours.

The forecast could not help with any of that. The forecast said 38.

Every good weather app answers a question about the atmosphere. Almost nobody is asking a question about the atmosphere. They are asking whether to put the rug on, whether the ground has dried enough to school on, whether the pavement will burn a dog’s feet at four o’clock, whether the rabbit will cope with tomorrow. The temperature is an input to that decision, not the decision.

So the work in this app is not weather. It is the arithmetic between the weather and the animal — and it turns out that arithmetic is genuinely hard, genuinely species-specific, and genuinely worth doing properly.

What "properly" turned out to mean

A few principles emerged early and then refused to bend.

Every number is visible, and every number is yours

There is no black box. Every constant behind every recommendation is listed in the app, with the reasoning for the value it holds — where it came from, and where it is this app’s own calibration rather than a published figure. If you disagree, you change it. If one horse in your barn runs cold whatever the chart says, you change it for that horse alone.

This matters because the alternative is asking people to trust a recommendation about an animal they know far better than we do. That is not a reasonable thing to ask.

Offline is the normal case

Barns have bad signal. Trailheads have worse. An app that needs a connection to tell you what it already knows is an app that fails exactly where animals live. So every advisory is computed on the device from a cached forecast, sunrise and moon phase are calculated rather than fetched, and place search works with the data shipped inside the app. A failed refresh shows you the last good answer with its age, never an error.

Hard stops are refusals, not low numbers

Lightning, ice, freezing rain, damaging wind, a creek running high — these do not produce a score of four out of a hundred that someone might talk themselves past on a day they really want to ride. They produce a refusal, in words. And the app will not offer you a "least bad option" when the only options left are those.

It says what it is not

It is not a severe-weather warning service, it is not veterinary advice, and it is not a flood authority. Those sentences appear on the screen where each recommendation is actually made — not folded into a settings page nobody opens. An app that gives confident instructions about a living animal owes people the edges of its competence, in the same breath as the instruction.

Nothing about your animals leaves your phone

Three things leave the device: a coordinate rounded to about a kilometre, a public river-gauge number if you pin one, and a species name with an image id to fetch a picture. There is no fourth item, no account, and no analytics in the binary. Your animals, your places, your thresholds and your history are on your phone and nowhere else.

This was not a marketing decision. It was easier to build an app with no servers holding user data than to build one that holds it responsibly, and it is much easier to promise.

The picture

There is a second thing this app does, and in practice it may be the more important one.

Every morning it shows you your kind of animal in the weather that is actually happening — correct sky, correct precipitation, correct time of day. A horse tail-to-the-weather in real rain. A labrador in snow. A cat under a porch roof while a shower clears.

Forecast accuracy is table stakes; every weather app has it and nobody switches for it. The picture is why anyone would switch. The advice is why they would stay.

The images are AI-generated and labelled as such in the app. Getting them to stop looking AI-generated took a great deal longer than expected, and taught us more about photography than about machine learning: rain is nearly invisible to a camera, so you show what it did rather than drawing it; a night scene cannot have both stars and a lit foreground; ground that has been stood in is churned, and smooth mud reads as fake instantly. The long-term intention is to replace them with real photographs of real animals, taken by the people who love them.

Who is making it

WeatherSpirit is built by Kyle — a small independent operation, not a venture-funded company with a growth team. There is no investor asking for engagement metrics, which is a large part of why the app has no engagement metrics.

The husbandry guidance behind the defaults is drawn from published extension and veterinary sources. Where a number is this app’s own calibration rather than a published figure — and several are, because no source publishes "+40°F for a body clip" — the app says so next to the number.

Where it is

In development, and not yet in either store. The core is built: the advisory engine for all four species, the offline forecast layer, place search, the profile editors, and a growing library of artwork. Still ahead: the home screen widget, the Apple Watch app and its complications, the voice shortcut, notifications, and the live river-gauge feed.

If you would like to know when it is out, leave an email address. One message, on launch day, and then the list is deleted.

Questions, corrections, or a threshold you think is wrong? Get in touch — particularly the last one.