The dataset

Query it in sixty seconds

Paste this into DuckDB — the desktop shell, or shell.duckdb.org in a browser. It reads the Parquet footer over HTTP range requests and pulls only the row groups it needs; it does not download the dataset. One URL per filing year, listed explicitly, because HTTP has no directory listing; the manifest carries the full list.

INSTALL httpfs; LOAD httpfs;

SELECT recipient_name_raw, recipient_state, amount_usd, tax_year, grant_purpose
FROM read_parquet(
  [ 'https://data.opengrants.io/funder-graph/2026.09.0/grants/filing_year=2023/part-0000.parquet' ],
  hive_partitioning = 1
)
WHERE funder_ein = '942278431'   -- The David and Lucile Packard Foundation
  AND amount_type = 'paid'
ORDER BY amount_usd DESC
LIMIT 25;

Reverse the question with WHERE recipient_ein_resolved = '363673599' AND match_confidence >= 0.90 to see who has funded Feeding America. The confidence filter is there for a reason; see the tiers.

The convenience JSON

Every funder page is also available as /api/funders/<ein>.json: the same payload the page renders from, CORS-open, no key. It is a convenience for people who want the page's numbers without the page. It is explicitly not an SLA-backed API; the Parquet is the stable interface.

Versions and citation

Versions are YYYY.MM.PATCH: the year and month of the IRS posting the release was built from, and a patch for a re-release that fixes a mapping without new source data. Every row carries object_id, tax_period_end and filing_submission_date, so any figure you publish traces to one filing.

Egeria Corporation. funder-graph: The Open 990 Funding Graph, dataset version 2026.09.0. Derived from IRS Form 990 and 990-PF e-file XML and the IRS Exempt Organizations Business Master File. Field mapping via the Nonprofit Open Data Collective IRS E-file Master Concordance File. https://github.com/egeria-corporation/funder-graph