Best of every model, in one forecast.
Built on data that no one else has.
WeatherMesh is our record-breaking AI model.
MetaMesh blends WeatherMesh with public AI and NWP models.
Tail events drive the biggest P&L swings. WindBorne models are designed to surface these early. For example, our model flagged a Storm Kristin 3 days before ECMWF and GFS converged..
WeatherMesh forecasts update every 20 minutes, replacing discrete 6-hour cycles with a continuous forecast signal. This enables tracking of signal evolution and confidence in near real time, aligned with intraday and day-ahead decision windows.
Proprietary observations, AI-native data assimilation, and high-frequency model cycles are designed to deliver best-in-class forecast accuracy across the full 15-day horizon.
# Install: pip install windborne
import windborne
# Auth via environment variables
# WB_CLIENT_ID and WB_API_KEY
# Fetch point forecast
forecast = windborne.get_point_forecast(
lat=40.71, lon=-74.01,
variables=["2t", "10u", "10v"]
)
# CLI also available
# $ windborne forecasts --helpEvery forecast product is delivered via REST API and an interactive insights platform. Your automated systems get programmatic access, and your team gets a meteorological-grade interface that shows why a model is producing a given forecast.
1-year backtests are available for WeatherMesh-5c, so you can score on your own variables and regions before going live. Our blended model MetaMesh dynamically recalibrates its weights as new input models arrive, so historical backtests would not accurately represent live model performance. For this reason, MetaMesh is available as a live forecast product only.
WeatherMesh-5c is our AI forecast model, trained on proprietary observations, running hourly. MetaMesh is our blended point forecast product: it combines HRRR, GFS, GFS Ensemble, IFS, IFS Ensemble, AIFS, AIFS Ensemble, and WeatherMesh models with dynamically learned weights that adapt by variable, lead time, region, and season. WeatherMesh is the raw AI signal. MetaMesh is the best-of-breed consensus. It updates on a rolling basis with hourly timesteps out to 15 days, optimized for over 5,349 METAR stations with global dynamic coverage.
REST API with JSON responses, plus a Python SDK. Most trading desks are pulling data within a day of onboarding. No portals, no file downloads.
For business and other inquiries, please use the form below.
For media inquiries, please contact press@windbornesystems.com.
WindBorne Systems is based in Redwood Shores, California. Our technology is invented, designed, and manufactured in the USA.
© 2026 by WindBorne Systems