Field note: asking an AI for a live river level, and where the answer comes from
WaterDataGPT answers with data fetched at the moment you ask, drawing on 621 keyless open APIs. Here is what happens behind one question.
- WaterDataGPT has 32 tools and a registry of 621 keyless open water APIs.
- Answers are built from live calls (USGS, UK Environment Agency, Open-Meteo), not from the model’s memory.
- Every answer links to the source it used, so you can check the number yourself.
Language models are fluent and often wrong about numbers. For water data, the fix is simple to state: do not let the model remember the river level, make it look it up.
WaterDataGPT does this with function calling. When you click a point on the map and ask about flow, the model picks one of 32 tools, calls the matching agency API from your browser, and writes the answer from what came back.
- United States: USGS NWIS gauges and NOAA forecasts.
- England: Environment Agency hydrology.
- Anywhere: Open-Meteo flood and precipitation, plus the 621-API registry and 1,939-dataset directory.