By default, Windy uses the ECMWF weather model. You can also change it to GFS or ICON, which many american forecast websites use.
ECMWF, GFS and ICON are made by national/international forecasting agencies. IBM also has its own proprietary weather forecasting service, notably used by The Weather Channel. Other apps mostly use one of these models or aggregate predictions from different services (e.g. AccuWeather claims to aggregate many different models, including those from national agencies around the world)
for people who live in europe, try AROME, it's very limited compared to the other models, but it does (IMO) give the best forecast.
in my experience AROME tends to forecast on the drier side of the weather. Since I'm in the UK I use AROME first and flick between ECMWF. (ECMWF tends to forecast on the wetter side!!)
for the next hour(s) forcast, I tend to use rain radar - windy has this, but it's not the best I use netweather.tv (there's many others though)
I'll add in a vote for the UK Met Office here too (particularly for the UK). AROME is a high resolution 'nested' model, where the region of interest is simulated at a higher resolution that the (often global) driving model. Similar to ICON-D2 (Germany [0])and UKV (UK [1]). These will typically produce a better downscale of the forecast to a local area.
Their utility depends a bit on what is causing the uncertainty in the forecast though. For the mid-latitutdes, a lot of the uncertainty comes from timing the arrival of a weather system. In these cases, the high resolution simulation doesn't help you, as it just incorporates the uncertainties of the lower resolution driving model.
The ensemble simulations that the weather servies perform will help you there, ECMWF has a really nice meteogram for individual locations [2].
IBM bought The Weather Company, which was the holding company of The Weather Channel. The forecasting technology is now owned by IBM, which licenses the weather data to The Weather Channel.
depends what you mean by 'rolling your own weather forecast'. If by that you mean you're going to do some statistical modelling on ensemble data, you're going to get a much better idea of what is likely to happen than you can from the average datasets you get on sites like windy
It's unclear to me if Windy is doing much more than creating a great presentation of existing data. Their website does list an open ML role, though..
I wonder how much rolling your own weather forecast is tantamount to rolling your own crypto??