The Atmos data pipeline

Atmos is built on three independent data sources that together produce a forecast that's both accurate and easy to read.

1. Open-Meteo for the numbers

Open-Meteo is a free, open weather API that provides forecast data on a 1km grid. Atmos pulls hourly and daily fields including temperature, precipitation probability, wind, gusts, dewpoint, cloud cover, UV index, and air-quality metrics. Because the data is hyperlocal, the prediction reflects your block, not your zip code.

2. RainViewer for the radar

RainViewer provides animated precipitation tiles with two hours of history and a 30–60 minute nowcast, plus a satellite cloud overlay (Atmos Pro). The radar timeline scrubber lets you see exactly when a front is going to clear your area.

3. gpt-4o-mini for the briefing

The numbers are the easy part. The hard part is answering "should I bring a jacket?" Atmos sends a structured prompt to gpt-4o-mini containing the next 12 hours of conditions, the user's saved alert preferences, and the current time of day. The model returns a short, plain-English briefing.

If the API is unreachable or the user is offline, Atmos falls back to a deterministic local summary so functionality is never blocked.

Why three sources, not one?

• Resilience: a single-source weather app dies when its provider has an outage.
• Specialization: Open-Meteo is best at structured forecasts, RainViewer is best at radar tiles, OpenAI is best at language.
• Cost: AI calls are only made when the user opens a relevant view, keeping per-user cost low.

If you want to dig deeper into how Atmos handles privacy and where data is stored, see the Privacy Policy.