6.59 million documents. 48 banks. One query.
Every hit is a receipt: doc ID + page + URL + legal class.
Free through 2026. Get pages, not stories.
Every declassified document, cross-indexed, cited, queryable by machine for a penny. The anti-hype move: no "we have the secrets" — we have the index of the index, and every answer carries a receipt.
"Oswald" returns pages from JFK and Cablegate and GIFiles and Epstein — 31 receipts in one call. Nobody else does cross-corpus.
Every hit = doc ID + page + URL + class (A official · B FOIA · WL WikiLeaks · ICIJ). The model can't summarize away the page. That's the citation layer.
FTS5 + JSON + x402. Agents don't scrape; they call it (free through 2026) and get structured answers. The tollbooth for the agent economy — a fair one, on public records.
WikiLeaks is not NARA. FOIA is not a leak. Every document knows what it is. That's the integrity product in a world of grifters.
WikiLeaks published · official NARA/DOJ/FOIA · ICIJ offshore graph · Epstein emails · BND inquiry. All FTS5, all labeled, all live.
Newest first. Every release is a separate bank, so a hit always tells you which dump it came from.
Already-public records · Doc ID or no claim · GET /v0/spookfiles/search?q= · payTo 0x53831fab…c7e0 · USDC on Base