What makes it different
Every number comes from a live API, the database or an official open dataset, and carries its source and fetch time. The language model writes wording only.
Probabilities, not adjectives
The share of 51 ECMWF and 31 NOAA GEFS ensemble members that cross crop-relevant thresholds, with a 90% interval, next to how often the IMD record (1991–2020) saw the same event in the same window.
Field evidence
Sentinel-2 NDVI and NDMI against the field's own 2017+ history, and Sentinel-1 radar when monsoon cloud blinds optical imagery.
Plans you can audit
Hybrid retrieval over the district's ICAR-CRIDA contingency plan. Every action cites a page, and a verifier rejects numbers that are not in the evidence and dates outside the forecast window.
A visible, self-correcting agent
A LangGraph agent with quality gates and source fallback chains, streamed live so you watch it plan, call tools, check them and fall back.
Proof without hindsight
Time Machine replays the forecast actually issued on a past date (from 2024-03-14) or, earlier, a LightGBM model on rain observed up to the day before.
Alerts where farmers are
FPO agronomists and KVK officers manage many fields in the web app; the farmers they serve receive Telegram alerts in Hindi or Punjabi.
Built on open data
- Open-Meteo ECMWF IFS and NOAA GEFS ensembles, ERA5-Land (CC BY 4.0)
- Copernicus Sentinel-2 and Sentinel-1 via Copernicus Data Space and Microsoft Planetary Computer
- IMD Pune 0.25° gridded daily rainfall
- NASA POWER daily point rainfall
- ICAR-CRIDA district agriculture contingency plans
- OpenStreetMap district boundaries and place search
Live status
Your browser asks the running API right now. If it cannot answer, this block says unavailable instead of showing a number.
API and model
Data sources the API has called
Demo districts
See it on a real field
Draw a field on the satellite basemap, pick the crop and sowing date, and watch the agent plan, call its tools, check their quality and verify its plan.