data.gov

data.gov is the US federal government's central open data portal, cataloguing more than 300,000 datasets from federal agencies, with links out to state, county, and city data too. Like its UK counterpart it is mostly an index rather than a store: each listing describes a dataset and points you to the files or the API that serves it. Coverage spans agriculture, climate, education, energy, finance, health, transport, and public safety.

The site is built on CKAN, so there is a documented catalogue API you can query for dataset metadata as JSON, filter by agency or tag, and page through results. The data files come in CSV, JSON, XML, and many other formats. Some of the most valuable holdings are not really files at all but standing APIs from agencies such as the Census Bureau, NOAA, and the Bureau of Labor Statistics, which you can hit for structured, current figures.

This is a strong base for any RAG system that needs authoritative US facts: census and demographic figures, weather and climate records, federal spending, environmental readings, and health statistics. The rich metadata on each listing helps a retrieval system understand what a dataset holds, and the sheer breadth means you can often find an official number rather than a second-hand quote of it.

Two things to keep in mind. Because so many agencies publish independently, quality and freshness vary, and datasets do occasionally disappear or move as portals are reorganised, so cache what you rely on. On licence, most US Government works are in the public domain, so you can reuse them freely with no attribution or share-alike obligation, but a minority of third-party datasets carry their own terms, so check the licence on each dataset before you ship it in a product. Treat it as the US sibling of data.gov.uk and data.europa.eu, and reach for the AWS Open Data Registry when you want very large scientific datasets hosted for cheap bulk access.