The whole sky, thirty days early.
Virga-1 is a foundation model of the atmosphere. It forecasts every square kilometre of Earth, every hour, a month ahead.
One model for the whole atmosphere.
Trained on eighty years of reanalysis and a decade of satellite radiances, Virga-1 rolls the planet forward an hour at a time: a 1 km grid, 37 levels of air, 120 variables.
Down to your street, not your county.
A global model sees a city as a handful of 9 km squares, and a storm cell can fall between them. Virga sees every kilometre. Move the lens across the squall line.
Under the lens
Virga · 1 km
0.0mm/h
Global · 9 km
0.0mm/h
1 kmgrid spacing, everywhere on Earth
Hourlysteps, out to day 30
64ensemble members, so every number has a spread
Ask for a window, not a forecast.
Tell Virga what the job can tolerate. It scores every hour of the next month by the chance it stays inside your limits.
Best windows
1+1 d · 18:00–00:0091%
2+3 d · 00:00–06:0088%
3+2 d · 18:00–00:0087%
Crane lift · Port of Rotterdam
241 of 720 hours pass · 3 windows
One sky, read four ways.
Hornsea wind farm · output
NEXT 72 H · MW
Output dips to 31% of capacity on Tuesday morning as the front stalls over the farm.
Capacity factor, next 7 days
47%
±6% across the ensemble
Graded every morning, in public.
Every run is scored against 11,402 weather stations and the following day’s analysis. This is last year’s card.
| Variable | Day 1 | Day 2 | Day 3 | Day 5 | Day 7 | Day 10 | Day 15 | Day 20 | Day 30 |
|---|---|---|---|---|---|---|---|---|---|
| Z500Geopotential, 500 hPa | −4 | −6 | −10 | −16 | −15 | −15 | −12 | −9 | −3 |
| T850Temperature, 850 hPa | −2 | −7 | −9 | −11 | −13 | −14 | −9 | −5 | −3 |
| T2MTemperature, 2 m | −4 | −5 | −6 | −10 | −12 | −10 | −7 | −6 | −3 |
| U10Wind, 10 m | −2 | −5 | −10 | −13 | −13 | −14 | −11 | −7 | −2 |
| MSLPSea-level pressure | −3 | −8 | −9 | −12 | −15 | −16 | −10 | −6 | −4 |
| TP24Rain, 24 h total | +1 | −3 | −5 | −9 | −10 | −8 | −6 | −6 | −2 |
| Q700Humidity, 700 hPa | −1 | −4 | −7 | −9 | −9 | −10 | −8 | −4 | −1 |
| TCCyclone track | +2 | +2 | −3 | −5 | −7 | −7 | −3 | −3 | −3 |
Wind, 10 m · day 3 · -10%
“Day-three wind is our whole winter. We cut imbalance costs by 18% in the first season.”
Rain, 24 h total · day 7 · -10%
“We plan harvest crews off the seven-day rain. Two lost days last year, against eleven the year before.”
Cyclone track · day 7 · -7%
“Track error at a week out is what moves our reserves. Virga gives us an extra two days of notice.”
One request. Thirty days of answers.
curl https://api.virga.example/v1/forecast \
-H "Authorization: Bearer $VIRGA_KEY" \
-d lat=51.92 -d lon=4.48 \
-d vars=t2m,gust,p_rain -d hours=48{ "point": { "lat": 51.92, "lon": 4.48, "name": "Rotterdam" }, "run": "2026-10-23T06:00Z", "model": "virga-1.4", "hourly": { "t2m": [10.5, 11.3, 12.1, …], "gust": [40, 45, 49, …], "p_rain": [0.08, 0.11, 0.16, …] }, "spread": { "t2m": [0.4, 0.4, 0.4, …] }}Rotterdam · next 48 h
Hover a field or a row
t2mTemperature, 2 m
10.5 → 8.3 °C
spread.t2mSpread
±0.4 → ±1.4 °C
gustGusts
peak 53 km/h
p_rainChance of rain
22 wet hours
Pay for the sky you cover.
Drag the circle over the places you care about, and its edge to size it. The ring it fits inside is your plan.
PLANET · beyond the outer ring
This week’s questions.
Seven days, seven things people ask before they sign up. Pick a day.
Mon · forecast
How good is a forecast thirty days out?
Honestly: good for patterns, not for hours. At a month out Virga tells you that the second week of November will run wetter and windier than normal along the North Sea coast, with a probability you can act on. It will not tell you it rains at 15:00 on the 19th, and it says so: the spread on every number widens with lead time.
Ask the sky.
Any place on Earth, any hour in the next thirty days. Try one here, then take a key and ask it a million times.
When
Lisbon · +7 d
Rain
High
7°
Rain
73%
Wind
23 km/h
Sample output · GET /v1/forecast?place=Lisbon