Standard Works: Epstein Files Search, OCR and AI Research Tools
Snapshot
| Resource | Standard Works |
|---|---|
| Operator | Efex AI, Inc. |
| Purpose | Document discovery, OCR and AI assisted review |
| Public resource | Epstein Files archive and OCR downloads |
| Access | Public archive advertised as free; commercial matter review priced separately |
| Reviewed | November 2, 2026 |
Standard Works is a document research platform offering search and AI assisted review. Its public Epstein archive can help researchers locate records, compare search results and return to document images. Its commercial eDiscovery service is a separate offering for customer document collections.
What the Public Epstein Archive Offers
The Epstein Files landing page advertises nearly 1.4 million files, more than 2.7 million OCR pages, filtering, sorting and original PDF viewing without a login. These figures describe the provider’s advertised collection, not an independently reconciled count of every government release.
The page supplies dataset downloads produced with Tesseract OCR. It describes higher accuracy OCR and audio and video transcripts as forthcoming. A dataset marked as still processing and the site’s accuracy disclaimer make it important to check the coverage of a particular search. The provider says most records came from DOJ and some from crowdsourcing.
Search and Analysis Tools
The product overview describes Boolean and semantic search, extracted quotations, AI analysis columns, OCR, corpus analytics, relationship graphs and agent assisted questions. Boolean search combines explicit terms; semantic search looks for related meaning. Researchers can use both approaches to find differently phrased material about the same subject.
The homepage identifies its example scores and evidence as unedited AI output. An extracted quotation should therefore be checked on the source page. A graph connection or model label should be traced to the specific record that produced it before it enters an article or report.
A Practical Research Workflow
Start with a narrow question, such as a date, place, transaction or exact document identifier. Record the terms used and the search date. Try spelling variations separately so that the results remain reproducible.
Open the underlying document rather than relying on a result snippet. Record its EFTA identifier when one is actually present, its physical PDF page, the displayed page number and its source URL. Check names, dates and amounts against the visible page. Keep a quotation distinct from a paraphrase and from your interpretation.
For correspondence, read the surrounding thread and any available attachments. Distinguish the date an event occurred from the date someone described it. For a financial record, identify the account, currency and transaction context before combining amounts. Do not treat duplicate copies as separate corroboration.
The Epstein File Tools provides other archives for comparison. A search that returns no result is a reason to examine spelling, OCR quality and collection coverage, rather than a conclusion that a record never existed. Data Privacy While Sleuthing covers precautions for handling personal information.
Costs and Commercial Access
The pricing page, reviewed November 2, 2026, lists hosting at $4.95 per GB per month, AI ingestion at $0.05 per document and additional prompts at $0.01 per document. Agent use is billed at Anthropic token rates. Long documents can count more than once.
The first case has a $5,000 ingestion minimum. For later cases, the FAQ says the company asks for at least $5,000 in ingestion annually; it also describes volume discounts and paid pilots. These published figures are not a quote for a particular collection. Confirm the document count, recurring storage, model usage, export costs and contractual commitments before commissioning a matter.
Privacy, Automation and Reuse
The Privacy and Terms page identifies Standard Works as a product of Efex AI, Inc. It distinguishes public account information from confidential paid customer material governed by a separate agreement. Public account queries and feedback may be used to improve retrieval, ranking and AI features.
The published terms restrict scraping, mirroring, redistribution and automated access, including AI agents. A publicly accessible archive is not permission to automate queries or republish a collection. Review the terms before using a download in another service. Organizational use also carries a provision permitting the company to identify the organization as a user.
Before uploading confidential survivor or attorney material, have the responsible legal team review the customer agreement, permitted uses, access controls, model providers, retention, deletion and export arrangements. Public access to released files does not make a separate confidential collection appropriate for upload.
Security Claims and Questions to Check
The Secure by Design statement describes required multifactor authentication, hosted software updates, vulnerability disclosure and customer audit logs. It describes enterprise SSO and SAML as roadmap work. These are the company’s stated controls, not an independent finding that every deployment meets a particular security standard.
For a professional evidence workflow, ask whether an export preserves original files, page references, email headers, attachments and metadata. Establish how document versions and duplicates are tracked. Test those requirements with nonconfidential sample records before relying on the service for a case.