JMessage: The Jmail Interface for Reading Jeffrey Epstein’s Text Messages

PostedAugust 5, 2026

UpdatedSeptember 14, 2026

ByEpsteinWiki

JMessage is an interactive tool that reformats text messages from the released Jeffrey Epstein records into an interface resembling Apple Messages.

The project is part of the larger Jmail collection created to make the Epstein files easier for the public to explore. Instead of presenting messages as raw database logs or difficult PDF pages, JMessage groups them by apparent participant and arranges them as familiar message conversations.

The interface is fast, readable, and useful for identifying conversational patterns. However, many participant names were originally redacted or anonymized. Some identities displayed by JMessage are estimates based on context, metadata, language patterns, artificial intelligence, and community research.

JMessage is therefore a discovery and navigation tool. Its participant labels and message attributions must be verified against the original evidence before publication.

Snapshot

Resource name: JMessage

Website: Jmail.world/messages

Resource type: Interactive text message archive

Parent project: Jmail

Primary developer: Jinglin Li, working with the Jmail project

Foundational data and processing: Michel de Cryptadamus Epstein Text Messages project

Primary sources: Text message records released by the House Committee on Oversight and Government Reform and later government releases

Interface style: Apple Messages inspired conversation view

Public access: Available without registration

Search features: Participant search and conversation selection

Known participants displayed: Steve Bannon, Melanie Walker, Anthony Scaramucci, Miroslav Lajčák, Larry Summers, Joi Ito, Stacey Plaskett, Arda Beskardes, Eva Dubin, Soon Yi Previn, Anil Ambani, and others

Critical limitation: Some participant identities are inferred and may be incorrect

What Is JMessage?

JMessage converts raw Epstein text message records into readable conversations.

The original government files were not presented as ordinary message threads. They appeared across thousands of pages and included technical fields, timestamps, telephone numbers, redactions, and fragmented conversations.

Messages involving the same apparent person could appear in multiple files. Participant names were frequently hidden. Researchers had to determine who sent each message, who received it, and how separate records fit together.

JMessage processes that material and groups the messages by apparent conversation partner. The interface displays each conversation as if the user were viewing Epstein’s message application.

This familiar design allows researchers to recognize exchanges, changes in subject, timing patterns, and the direction of a conversation more easily than they could when reading the original logs.

Who Built JMessage?

Software developer Jinglin Li helped build JMessage as a contribution to the larger Jmail project.

In a January 2026 account of the development process, Li explained that JMessage began with more than 20,000 pages of released PDF records.

Li used the open source work of Michel de Cryptadamus as the project’s foundation. That earlier project had already downloaded records, extracted text through optical character recognition, classified documents, identified message files, and reformatted many conversations.

The underlying work is publicly available through the Michel de Cryptadamus GitHub repository.

Li then wrote software to group messages by apparent participant and timestamp and convert the organized data into a structured format for the JMessage interface.

JMessage shows how independent research projects can build upon one another. Government files became an open source processed archive. That archive then became an interactive public interface.

How the Interface Works

The JMessage homepage presents a list of conversations ordered by recent activity.

Each entry displays an apparent participant, the date of the most recent visible message, and a preview of the conversation.

Researchers can select a person and scroll through the available message history. The interface places Epstein’s messages and the other participant’s messages on opposite sides of the conversation.

Dates and times divide the exchanges. Links, partial attachment descriptions, and extracted text appear within the message stream.

The interface also includes search functionality. A researcher can search for a participant rather than scrolling through every available conversation.

JMessage connects with the wider Jmail collection, including Jmail, JPhotos, JDrive, JFlights, JWiki, Jamazon, Jotify, and other interfaces.

Known and Tentative Participant Labels

One of the most responsible features of JMessage is its use of tentative labels.

Some conversations are displayed with confirmed looking names. Others are labeled “Maybe” and include a question mark.

Examples visible in the interface include “Maybe: Duda Entourage,” “Maybe: Celina Dubin,” and “Maybe: Michael Wolff.”

These labels acknowledge that the original release did not always identify the participant. The displayed name represents a conclusion or best estimate rather than confirmed metadata.

Researchers must preserve that uncertainty. A tentative JMessage label should never be converted into a definitive attribution merely because the interface makes the conversation look complete.

If JMessage says “Maybe,” any article referencing the conversation must also state that the identification is uncertain.

How Participants Were Identified

According to Jinglin Li, participant identities were reconstructed through several methods.

Researchers examined conversational language, dates, locations, telephone metadata, events mentioned in the messages, writing style, artificial intelligence analysis, and suggestions from members of the public.

This method can produce strong identifications when several independent clues point to the same person.

It can also produce mistakes.

A person may share a writing style with someone else. A location can fit several people. A telephone may be used by an assistant. A message may refer to an event involving someone who was not the participant. Artificial intelligence may treat a plausible identification as a certain one.

Li added a warning to the interface explaining that attributions may be incorrect. The project also accepts corrections from users.

The Importance of Message Direction

Raw message logs can make it difficult to determine who wrote a particular statement.

Michel de Cryptadamus, whose data helped support JMessage, publicly acknowledged initially confusing messages sent by Steve Bannon with messages sent by Epstein.

That mistake is especially understandable when names are redacted and the records appear as technical logs instead of conversations.

JMessage reduces this risk by placing each apparent participant on a separate side of the interface.

The improvement is significant, but the visual layout depends on the accuracy of the underlying parsing. If a source field was misread, a visually convincing conversation could still reverse the speakers.

Researchers must confirm the sender and recipient in the original government file before publishing a quotation.

The Steve Bannon Conversation

Steve Bannon appears prominently in JMessage.

The visible conversation list shows messages attributed to Bannon continuing through July 6, 2019, the day Epstein was arrested on federal sex trafficking charges.

The larger Michel de Cryptadamus analysis reported that Bannon accounted for approximately 1,255 of the 2,108 incoming messages identified in the November 2025 archive. That would make him the most frequent apparent correspondent in that particular released collection.

This figure should be interpreted carefully. It describes the available message release, not Epstein’s complete lifetime communications.

The number may reveal as much about the documents selected and preserved by Epstein’s estate as it does about Epstein’s entire messaging history.

The messages are nevertheless important because they document sustained communication concerning politics, media, international figures, business plans, public narratives, cryptocurrency, and Bannon’s activities after leaving the White House.

Message frequency does not establish criminal conduct. Each significant exchange must be examined on its own terms and within its original evidentiary context.

The Melanie Walker Conversation

Melanie Walker appears as another prominent participant in the archive.

The conversation list shows messages attributed to Walker continuing into February 2019. The earlier source analysis counted approximately 340 incoming messages attributed to her.

Messages associated with Walker concern science, health, technology, philanthropy, investments, travel, and people within Epstein’s professional network.

As with every participant, the number of messages does not establish knowledge of Epstein’s crimes. The content, timing, context, and independent evidence determine what a conversation actually demonstrates.

Researchers should distinguish professional, social, logistical, financial, and potentially criminal communications rather than treating every message as equivalent.

Other Conversations in the Interface

The JMessage homepage displays conversations attributed to a wide range of public and private figures.

Visible names include Larry Summers, Joi Ito, Stacey Plaskett, Arda Beskardes, Miroslav Lajčák, Anthony Scaramucci, Eva Dubin, Terje Rød Larsen, Soon Yi Previn, and Anil Ambani.

Some names are presented tentatively.

The presence of a person in the interface means that the project associates that person with a released conversation. It does not prove criminal participation, knowledge of abuse, or even that every message in the thread was correctly assigned.

Researchers should also consider whether a name is associated with the telephone subscriber, the person using the device, or an identity inferred from conversation content.

Example: The Tentatively Identified Celina Dubin Conversation

JMessage includes a conversation labeled Maybe: Celina Dubin.

The page describes the apparent participant as a daughter of Glenn Dubin and a childhood acquaintance within Epstein’s social circle. It displays 2019 messages concerning birthdays, family, travel, financial gifts, credit history, news coverage, and Epstein related publicity.

The page’s tentative label is critical. It tells readers that the participant identification was reconstructed rather than confirmed through an unredacted name in the source.

Even when the content appears consistent with a particular person’s biography, consistency is not the same as authentication.

Researchers should not quote these messages as communications with Celina Dubin without explaining the attribution uncertainty and locating corroborating evidence.

For broader context, see EpsteinWiki’s investigation of Epstein’s social events and personal network and its guide to mapping the Epstein network.

Connections to Original EFTA Evidence

JMessage was built from processed versions of government released message records. Corresponding EFTA files can be examined through Epstein Data.

Examples of evidence files associated with the underlying reformatted message collection include:

The original files should be used to confirm the wording, date, message direction, metadata, and surrounding conversation.

Attachments Are Frequently Missing or Unidentified

Some JMessage entries state that a message contained an attachment, probably a photograph, video, or animated image.

The interface may not display the attachment itself. In those cases, the reader does not know what the recipient actually received.

A message responding to an unseen photograph can be misunderstood when separated from the attachment. A short response such as “yes,” “wow,” or “who is that” may have little evidentiary value without the missing image.

Researchers should not assume what an attachment depicted. If the attachment is unavailable, the limitation should be stated explicitly.

The broader JPhotos archive may help locate visual material, but a photograph must be connected to a specific message through metadata or another reliable evidentiary link.

Links and Extracted Text May Contain Errors

JMessage displays links shared within conversations. Some links contain spacing errors, broken parameters, or characters altered during text extraction.

These problems can result from optical character recognition, damaged source files, formatting conversions, or the original message export.

A broken link can sometimes be reconstructed from its visible domain and path. Researchers should avoid altering a link and then presenting the reconstruction as the exact original without explanation.

Spelling errors and unusual symbols may also come from the source record, text extraction, emoji conversion, or device encoding.

Quoted wording should be checked against the original file image whenever possible.

Artificial Intelligence Assisted the Identification Process

Jinglin Li disclosed that large language models were among the tools used to identify redacted participants.

Artificial intelligence can compare language, dates, places, relationships, and known events across a large archive. That makes it useful for developing possible identifications.

It cannot authenticate a person.

A model may select the most plausible candidate even when the available evidence is insufficient. It may also reinforce an incorrect assumption supplied by earlier researchers.

Artificial intelligence output should be treated as a lead. Confirmation requires metadata, telephone records, unredacted duplicates, independent communications, testimony, or other evidence.

Community Corrections and Crowdsourcing

JMessage has received corrections and tips from users.

Crowdsourcing can improve the archive when a researcher locates an unredacted duplicate, recognizes a telephone number, identifies an event mentioned in a conversation, or provides a stronger explanation for an attribution.

Crowdsourcing can also introduce confident but unsupported identifications.

A correction should be evaluated according to the evidence supporting it, not the number of people repeating it.

The strongest correction includes the relevant document, metadata, unredacted copy, or independent record. A social media consensus is not authentication.

Privacy and Redaction Concerns

Li reported finding telephone numbers and other sensitive information that had not been properly redacted in the released documents.

Some government files used visual redaction that left the underlying text accessible through copying and pasting.

The ability to recover hidden text does not automatically make its publication ethical.

Researchers should not republish survivor names, private telephone numbers, home addresses, financial account details, medical information, or identifying information concerning people who are not legitimate subjects of public investigation.

EpsteinWiki’s mission requires transparency and evidence. It also requires protecting survivors and unrelated private individuals from further harm.

The Interface Is Not the Evidence

JMessage is a transformed representation of government records.

The transformation includes text extraction, document classification, grouping, participant attribution, timestamp sorting, formatting, and visual presentation.

Every one of those steps can improve accessibility. Every step can also introduce an error.

The original government file remains the primary evidence. The JMessage page documents how the project interpreted and displayed that evidence.

Researchers should cite JMessage for its contribution to accessibility or identification while citing the original EFTA record for the factual claim.

How Researchers Should Use JMessage

Begin by selecting a conversation from the JMessage directory.

Check whether the participant name is presented as confirmed or tentative.

Read the entire available conversation rather than isolating one dramatic sentence.

Record the date, time, message wording, apparent sender, and apparent recipient.

Locate the corresponding government record through the original Michel de Cryptadamus dataset, JDrive, or Epstein Data.

Compare the interface with the source file. Confirm that the visual sides of the conversation match the underlying sender and recipient fields.

Search for duplicate copies that may contain clearer metadata or fewer redactions.

If the identity remains uncertain, preserve that uncertainty in any published article.

Relationship to Jmail, Epstein Data, and EpsteinWiki

JMessage is one part of the larger Jmail research environment.

Jmail presents Epstein’s released emails through an inbox interface.

JWiki uses artificial intelligence to create encyclopedia style profiles from the archive.

JDrive provides document browsing and text searching.

JMessage presents extracted text conversations in a familiar format.

Epstein Data provides direct evidence pages and searchable records.

EpsteinWiki creates structured, evidence based articles explaining the people, institutions, properties, financial systems, and investigations connected to the files.

JMessage can reveal the conversation. Epstein Data can provide the evidence. EpsteinWiki can explain the documented context.

Key Takeaways

JMessage converts raw Epstein text message logs into a familiar conversation interface.
Jinglin Li helped build JMessage as part of the Jmail collection.
The project relies heavily on the open source document processing work of Michel de Cryptadamus.
Messages are grouped by apparent participant and timestamp.
Many participant names were originally redacted or anonymized.
Some identities were reconstructed using context, metadata, writing style, artificial intelligence, and community tips.
The word “Maybe” identifies participant labels that remain uncertain.
The visual interface can improve message direction but does not independently authenticate the sender.
Missing attachments and extraction errors can change how a message appears.
Every important quotation and attribution should be verified against the original EFTA record.

Why JMessage Matters

Public records are not truly accessible when only technical experts can interpret them.

JMessage turns fragmented logs into conversations that ordinary readers can follow. It exposes timing, repetition, familiarity, and conversational context that can disappear inside raw data.

Its design also demonstrates the value of collaboration. Michel de Cryptadamus processed and organized the records. Jinglin Li transformed that work into an interactive interface. Jmail connected it with a larger public research ecosystem.

The project’s usefulness depends on preserving the distinction between readability and authentication.

JMessage makes the conversations easier to see. Researchers must still prove who was speaking.

Sources

Tags:

Communication Records