KORPEX Request a demo
Korpex for Email Available now

Narrative intelligence for large-scale email review.

Two million messages. A dozen custodians. One story. Korpex rebuilds every conversation in the collection and reads it for what people knew, meant, and decided, then hands you the findings, each linked to its source.

Request a demo What Korpex reads for
Custodian narrative
Matter: Q2 valve shipment · 2.1M messages
RC
R. Chen
Operations · 3 mailboxes · Mar 2 – Jun 30

On March 28, Chen received QA's second notice that the valve batch had failed pressure testing KX-1427. The next morning he proposed taking the discussion off email KX-1431. No further written discussion of the test results appears in any custodian's mailbox for the following five days.

On April 3, in response to the VP's question about the ship date, Chen wrote "We're good. Shipping on the 12th" KX-1502. The same day, QA re-sent the failed test report to Chen's second mailbox KX-1509.

Knowledge Channel change Contradiction 4 citations
Illustrative matter. Names, IDs, and messages are fictional.
Finds the circumstantial evidence keyword search misses.
Korpex reads for implication, timing, and tone: the warning that went unanswered, the meeting moved off email, the reassurance that contradicted what the sender had just been told.
Follows the story across every thread and every custodian.
A narrative rarely lives in one inbox. Korpex tracks it as it moves between people, forwards, and subject lines, and builds the chronology as it goes.
Built for massive productions, yours and opposing counsel's.
Korpex ingests the other side's document dump as readily as your own collection. Their volume stops being a burden and becomes your advantage.
What Korpex reads for

Signals, not keywords.

Every conversation is read for the things a litigator would underline. Each signal is a finding with citations, not a score.

Knowledge
Who knew what, and when they knew it. Including knowledge that arrived by forward, attachment, or a reply they never answered.
Warnings raised and dropped
The flag that was raised, acknowledged, and never acted on. Korpex pairs each warning with what happened next, or didn't.
Decisions and instructions
Who decided, who was told, and what they were told to do. The chain of command as it actually ran, not as the org chart says.
Channel changes
"Let's take this offline." "Call me." The moment a written record goes quiet, and the gap that follows it.
Contradictions
What was said internally, set against what was said to customers, regulators, auditors, or the board, in the same week.
Tone shifts
The point where a thread turns careful, formal, or silent. Often the first sign that someone realized what the record would show.
From ingest to production

Fits the workflow you already run.

Korpex sits between collection and review. It takes what your vendor or the other side delivers, and hands your review platform work product it can use.

Ingest
Everything the matter produced
  • PST, OST, MSG, EML, and MBOX
  • Microsoft 365 and Google Workspace exports
  • Productions from opposing counsel, native or with load files
  • Attachments, embedded images, and scanned pages (OCR)
Process
One corpus, read as conversations
  • Threading across replies, forwards, and renamed subjects
  • Exact and near-duplicate suppression
  • One identity per person across mailboxes, aliases, and signatures
  • Dates normalized across time zones and clients
Deliver
Work product your team can use
  • Findings with citations, by issue and by custodian
  • Chronologies and custodian narratives
  • Potentially privileged threads flagged for counsel's review
  • Tags and work product exported as standard load files to your review platform
Questions litigators ask

Straight answers.

Everything your partners and your lit-support team will want to know before a pilot.

Is it defensible?
Every finding cites its source messages, every step is logged, and results are reproducible on the same corpus. Korpex is a tool for counsel's review, not a substitute for it.
Where does my client's data go?
Nowhere. Korpex runs in your environment or in a single-tenant deployment you control. Your data is never used to train models.
Does it replace document review?
No. It reads first and points your reviewers at what matters, so review hours go to the documents that decide the case instead of the ones that merely match a term.
What about privilege?
Korpex flags threads that show signs of privilege for counsel's review. It never makes a privilege call on its own, and nothing is produced without a reviewer's sign-off.
How does it handle the other side's production?
The same way it handles yours. Their production is ingested, threaded against your own collection, and read as one corpus, so the gaps between what they said and what you have are visible.

Bring a production. Leave with findings.

A pilot on one matter, under NDA, in your environment if you prefer.

Request a demo