Questions and answers
The short version of what Nimbo does, what it does not, and what happens when something goes wrong.
The forecast
Where does Nimbo forecast?
Most of Europe, wherever European weather radar reaches. Italy is seen only from far away, by its neighbours’ radars, and a few edges of Portugal and Spain drop out on some days. The coverage page has the map.
Does the public map show the same forecast as the API?
What happens when the radar feed is down?
Usually, nothing you would notice. The radar reaches us by more than one route, and when the main one stops, a backup takes over by itself.
When the radar itself stops, there is nothing new to forecast from, and Nimbo says so rather than guess:
- every API answer says how old its data is, and sets
stale: trueafter 18 minutes; - after 30 minutes, answers carry a
notice, the map shows a banner and the status page shows the outage; - after 2 hours, the API answers
503 forecast_unavailableinstead of serving an old forecast as if it were new.
How accurate is it?
The first version of Nimbo’s model went live on 23 October 2026. It had studied 64 days of rain over Spain and Portugal, and that was all it knew.
So we gave it more rain. Then France. Then the rest of Europe. Some versions lost to the one before them and never left the lab; the winners went live. The model running today is version 11, and it has learned from 704 days of European rain.
Like every nowcast, it is sharpest in the first hour and softer towards three. How it does against the open-source standard is on the scoreboard, in public.
Does it forecast snow?
Not yet. Nimbo does not tell rain and snow apart: every value is reported as rain (precip_type: "rain").
Why shouldn’t I vibe-code this myself?
You can. The radar data is out there, good nowcasting models are published, and an afternoon with an AI assistant will get you rain moving across a map. Then the fun starts.
Getting the motion right is the hard part: rain that stutters, smears or drifts the wrong way looks fine in a demo and wrong on a real afternoon. It is where most of our time went. Training costs real money too: our rented GPUs alone came to around $100, and that was after we learned to rent cheap ones. And once it works, it has to keep working: pulling radar every few minutes, through outages, with an ingestion pipeline that does not fall over at 3 a.m., costs at least about $50 a month before anyone calls it.
Or you could use ours. The free plan is 20,000 calls a month.
Plans and keys
What can I do with the free plan?
Anything, including commercial use, as long as you show the attribution (see the API docs). It includes 20,000 calls a month and needs no card.
What happens when I reach my monthly allowance?
On the Free plan, calls stop at 20,000 with a 429 until the month ends. Paid plans keep working for 10 % more, then stop the same way. There is no overage billing, so no surprise bills. You get an email at 80 % and at 100 % of your allowance.
How are map tiles counted?
A tile our server draws for you is one call. A tile answered from our cache is not counted, nor is the latest redirect. The pricing page turns this into map views.
Can I get a refund?
Yes: within 14 days of paying, with no questions asked. After that you can cancel at any time from your dashboard, and you are not charged again; past months are not refunded.
Support
How do I get help?
Write to [email protected]. We reply within 2 working days.
Is there an uptime guarantee?
No. Nimbo is a best-effort service, and says so. Its history is public on the status page: whether the API answered, and whether the forecast was fresh.
Will the API change under me?
Not without warning. v1 is stable: fields may be added, but none is renamed or removed without 6 months’ notice by email.