Frequently asked questions about the data, search, and features.
A search engine for the Epstein Files — approximately 1.2 million documents released by the U.S. Department of Justice as part of investigations related to Jeffrey Epstein. The original PDFs are published at justice.gov/epstein. This tool lets you search all of them by keyword, date, and other filters — and links back to the original PDF for every result.
The DOJ published the documents in 12 numbered batches. The text used by this site was extracted from each PDF and structured by Tommy Carstensen's Epstein Files archive. Each document on this site links back to the source PDF on justice.gov.
About 313 documents from later batches were quietly removed from the DOJ website after they were first published. Their PDFs are no longer available on justice.gov. This site keeps those records searchable and flags them with a removed from DOJ badge so you know the original file is gone. You can filter search results to show only removed documents using the DOJ Status filter. The removal tracking comes from Tommy Carstensen's deleted files list.
Type any keyword or phrase in the search bar. Results are ranked by relevance across the full text of all 1.2 million documents. Leaving the box empty shows all documents. Every search is captured in the URL, so you can bookmark or share any search.
Filters
Sort options
Dates are pulled automatically from the text of each document — scanning for common date formats near the top of the file, where email headers, letter headings, and filing stamps typically appear. This works for roughly 87% of documents. The rest are labelled undated because no recognizable date appeared in the text. Undated documents are still fully searchable but are excluded from date-based sorting and the timeline.
Date extraction isn't perfect. Some documents carry incorrect dates due to scanning errors, or because the first date in the text was a reference (e.g. “per the 1966 Freedom of Information Act”) rather than the document's own date. Treat dates as a useful guide, not a guarantee.
The homepage shows a year-by-year activity chart for the entire corpus. Each column of squares represents one week; darker squares mean more documents on that day. Click any square to jump to documents from that date. Amber dots mark the globally busiest days — the top 1% by document count.
The same calendar appears in the search sidebar, so you can combine a date filter with any active keyword search. There's also a “jump to date” input for going directly to a known date, and prev/next arrows to step through days that actually have documents.
Names of people, organizations, and places mentioned across the documents are automatically identified and counted. Browse all entities or click any name in a search result to see every document that mentions it. Each entity page also shows a network graph of other names that frequently appear in the same documents.
Entity recognition is automatic and imperfect — scanning errors and ambiguous names introduce some noise.
On any document page you can ask Claude to summarize the document or answer specific questions about it. To use this, you need your own Anthropic API key, which you enter in Settings. The key is stored only in your browser — it is never sent to or stored by this website. You pay Anthropic directly for any usage. Three models are available: Haiku (fast, low cost), Sonnet (balanced), and Opus (most thorough), switchable from the AI panel on any document page.