EpsteinGraph
EpsteinGraph is an independent searchable archive and discovery platform for records connected to Jeffrey Epstein. It combines document search, artificial intelligence generated summaries, entity extraction, media transcription, flight log tools, timelines, and graph based exploration. The platform can help researchers locate relevant records, but its automated outputs are research leads rather than verified findings.
Snapshot
Name: EpsteinGraph
Website: EpsteinGraph
Type: Independent document archive and research discovery platform
Operator: The site identifies its operator only as an independent developer
Primary purpose: Searching, organizing, transcribing, and connecting public records related to Jeffrey Epstein
Main source collections claimed by the platform: Department of Justice releases under the Epstein Files Transparency Act, House Oversight materials, Estate of Jeffrey Epstein records, and court records
Important limitation: Entity names, summaries, transcripts, dates, and graph connections may be created or extracted by artificial intelligence and require confirmation in the underlying record
Access: Publicly accessible without an account
Overview
EpsteinGraph describes itself as a research archive built for journalists, researchers, investigators, and members of the public. Its central purpose is to make a very large body of Epstein related material easier to search and connect than traditional government document portals.
The site is not a government archive, court repository, news organization, or substitute for the original record. It functions as a discovery layer. Researchers can search documents, people, subjects, and locations, then use the results to identify records that should be examined through the Department of Justice Epstein Library, Epstein Data, court dockets, congressional releases, or another primary source.
EpsteinGraph states that no editorial judgment is made about the contents of the documents it indexes. That distinction matters. A name appearing in a search result, entity page, relationship graph, transcript, or list of prominent figures does not establish misconduct, participation in a crime, or even a personal relationship with Epstein.
What EpsteinGraph Provides
The platform offers full text searching across documents and extracted text. Its homepage allows users to search for names, places, organizations, subjects, and other terms. Records may also be explored through document type collections that include photographs, videos, audio, flight logs, emails, travel documents, financial records, and legal filings.
EpsteinGraph also provides person pages and graph based relationship tools. These features are designed to show where names or entities occur together across indexed materials. The platform includes timeline data, lists of frequently mentioned people, popular document views, and a dedicated flight log search covering routes, passengers, aircraft, and dates.
Audio and video files are transcribed so that spoken material can be searched. The site also states that it uses artificial intelligence to extract text from handwritten records that ordinary optical character recognition may not capture.
According to the platform’s Best Practices page, its indexed data included 1,321,030 documents, 2,291 videos, 152 audio files, and approximately 230,000 extracted entities when that page was reviewed. These are self reported and changing platform totals. They should be treated as a description of the archive at a particular time, not as an independently audited count.
Document Sources Identified by the Platform
The EpsteinGraph About page identifies four broad source groups:
- Records released by the Department of Justice under the Epstein Files Transparency Act
- Documents and hearing materials released by the House Committee on Oversight and Government Reform
- Records associated with proceedings involving the Estate of Jeffrey Epstein
- Court transcripts, filings, and evidence associated with Epstein related litigation
Researchers should trace each result to its original source. The Department of Justice Epstein Library is the official federal portal for material released under the Epstein Files Transparency Act. Epstein Data provides searchable access to EFTA records and direct document identifiers. Federal court records should be checked through PACER or CourtListener, while congressional material should be compared with the House Oversight Epstein records.
How the Graph Connections Work
EpsteinGraph states that its network relationships are created from document cooccurrence. In practical terms, two names may be connected because both were extracted from the same document or document group. That does not necessarily mean the people communicated, met, traveled together, conducted business together, or knew one another.
Cooccurrence can be useful because it exposes clusters that may deserve closer examination. It can also be misleading when a filing, news clipping, index, witness statement, contact list, or investigative memorandum mentions many unrelated people. A court filing may name attorneys, witnesses, public officials, companies, and people discussed only as background. An automated graph may display those names as connected even though the document does not establish a substantive relationship.
The platform expressly warns users not to treat cooccurrence as causation, frequency as evidence, or graph connections as proof of wrongdoing. EpsteinWiki applies the same standard. Every apparent relationship must be classified according to what the underlying record actually shows.
Artificial Intelligence and Extraction Limits
EpsteinGraph uses artificial intelligence for summaries, entity recognition, transcription, handwriting extraction, dates, and relationship mapping. These tools can make difficult records discoverable, but each layer can introduce errors.
Entity extraction may merge different people with similar names, split one person into multiple profiles, mistake ordinary words for names, or connect a name to the wrong identity. Transcripts may mishear speakers, names, dates, and technical language. Handwriting recognition may convert uncertain marks into confident text. Date extraction may identify a filing date, upload date, email date, or date mentioned in the record without distinguishing among them. Summaries may omit qualifiers, denials, procedural context, or uncertainty contained in the source.
For these reasons, EpsteinGraph’s own research guidance calls the platform a research tool rather than a truth engine. Its artificial intelligence outputs should never be quoted as if they were sworn testimony, authenticated evidence, a judicial finding, or a verified transcript.
Recommended Research Workflow
Researchers should begin with a narrow search using an exact name, spelling variation, email address, organization, location, date, or known document identifier. Results should then be opened individually rather than evaluated only through the graph or search preview.
The next step is to identify the original document number, release collection, file name, page number, and source agency. When an EFTA number is available, the record should be located through Epstein Data and compared with the official DOJ release. Court filings should be checked against the docket. House materials should be confirmed through the committee’s official release.
Researchers should read the pages before and after a match. A search result may capture a name without the sentence that explains why it appears. The surrounding pages may contain a denial, quotation, exhibit label, attorney argument, or attribution that completely changes the meaning.
Any claimed relationship should be corroborated across independent records when possible. Researchers should distinguish a direct communication from a third party mention, a passenger entry from a proposed itinerary, an address book listing from a documented meeting, an allegation from a finding, and a machine generated transcript from an authenticated transcript.
Anonymous Uploads and Provenance
EpsteinGraph invites users to submit files and permits anonymous submissions. This may help preserve records that disappear from government servers or become difficult to locate. It also creates an additional verification obligation.
A user submitted file should not automatically be treated as authentic merely because it appears on the platform. Researchers should document where the file originated, who released it, whether it matches an official copy, whether pages are missing, whether the file has been altered, and whether its metadata or numbering can be independently confirmed.
For EpsteinWiki purposes, an anonymously submitted record should remain a lead until its provenance is established. If no official or independently authenticated copy can be located, the limitation should be stated plainly and the record should not be used to make accusations about an identifiable person.
Privacy and User Data
The EpsteinGraph Privacy Policy states that users do not need an account and that the site does not use cookies, sell personal data, or operate third party advertising. It says starred documents are stored locally in the user’s browser.
The policy also states that its analytics service receives page views, referring pages, IP addresses, approximate locations, device and browser information, time spent on pages, and a random visitor identifier stored in local browser storage. The site therefore collects limited technical and usage information even though it does not require user registration.
Researchers handling sensitive inquiries should consider this disclosure before searching for survivor names, protected identities, confidential leads, or unpublished investigative terms.
Evidence Appearances
EpsteinGraph is a later research platform that indexes records. It is not itself an EFTA evidence record, court exhibit, witness statement, or investigative finding. No specific EFTA document appearance is necessary to establish the existence or functions of the website.
When EpsteinGraph leads to an EFTA record, EpsteinWiki citations should link to the exact evidence item through Epstein Data rather than citing only the EpsteinGraph summary, entity page, or search result.
What EpsteinGraph Does Not Establish
An appearance on EpsteinGraph does not establish that a person knew Jeffrey Epstein.
A line on its relationship graph does not establish direct contact or collaboration.
A high mention count does not establish importance, culpability, or investigative relevance.
An artificial intelligence summary does not establish what a document legally proves.
An extracted date does not necessarily identify when the underlying event occurred.
A transcript does not replace listening to the original recording.
An indexed file does not automatically establish authenticity, completeness, or chain of custody.
Editorial Assessment
EpsteinGraph can be valuable for finding names, documents, media, and possible connections across a collection too large for ordinary manual review. Its strongest use is discovery. Its greatest risk is that an automated graph or summary can make a weak, incidental, or mistaken connection look more certain than the underlying evidence supports.
EpsteinWiki may cite EpsteinGraph as a research tool or discovery source, but factual claims about people, events, communications, travel, money, allegations, and legal outcomes should be cited to the underlying primary record. The platform’s own public guidance supports that approach.
Related EpsteinWiki Pages
Department of Justice Epstein Library
Questions for Researchers
- Which artificial intelligence models and extraction procedures are used for summaries, transcripts, handwriting recognition, entity identification, and date extraction?
- How does the platform distinguish people who share the same or similar names?
- Can every indexed item be traced to an original agency release, court docket, congressional production, or documented estate source?
- How are corrected transcripts, mistaken identities, duplicate documents, and removed records handled?
- Does the platform preserve file hashes, release paths, upload dates, and other chain of custody information?
- How are anonymous submissions authenticated before they are incorporated into search results or relationship graphs?
- Are archived versions of changing entity pages, graphs, summaries, and platform statistics preserved for later review?
- What safeguards prevent survivor names, protected identities, and sensitive personal information from being amplified through automated extraction?