Foxxception Launches WeatherLens, a Weather API With Its Own Machine-Learning Forecast Correction

JACKSONVILLE, Fla. (April 2, 2026) Foxxception LLC today announced the launch of WeatherLens, a developer weather API that does more than pass along someone else's forecast. WeatherLens combines forecasts from the National Weather Service and Open-Meteo, then runs them through its own machine-learning models to correct the errors those forecasts tend to make.

Every weather source is wrong in predictable ways: running warm in one climate, missing rain in another, losing skill several days out. WeatherLens learns those patterns. It compares past forecasts with what actually happened at hundreds of locations around the world, and uses gradient-boosted (XGBoost) models to correct each new forecast's high and low temperatures and rain probability before it reaches the developer.

How it works

• Adaptive blending: NWS and Open-Meteo forecasts are weighted by each source's track record for that measurement and lead time, and every forecast comes with a confidence score.
• Machine-learning correction: models trained on tens of thousands of forecast-versus-actual pairs, using factors such as elevation, climate zone, terrain, distance to the coast, season and how far ahead the forecast looks.
• Retrained every week as new outcomes come in, with a guard that blocks any model that performs worse than the one before it.
• Measured, not assumed: an accuracy endpoint reports how each source has actually performed by month and lead time.

In walk-forward backtesting on more than 91,000 forecast-versus-actual pairs, the correction models cut average high- and low-temperature error roughly in half, from about 2.0°F to about 0.95°F.

Built for weather-dependent decisions

• Travel and outdoor apps: 16-day forecasts paired with multi-year climate baselines, so travelers know whether a destination is usually rainy that time of year.
• Agriculture: climate-aware forecasts for planting windows, irrigation scheduling, frost alerts and harvest timing.
• Transport and logistics: forecasts along every waypoint of a route before dispatch, with confidence scores that flag the uncertain segments.

Four endpoints

/forecast (up to 16 days), /climate (multi-year weighted climate normals), /accuracy (source accuracy by month and lead time) and /historical (up to 365 days).

"Weather apps all pull from the same few models, so they're all wrong in the same ways," said Jeremy Foxx, founder of Foxxception. "I wanted an API that learns where those forecasts miss and fixes it, for the people who plan trips, crops and deliveries around the weather."

Availability

WeatherLens is available now at weatherlens.dev. The Free plan includes 100 requests a day; Starter is $9 a month and Pro, which adds historical data, is $29 a month.