For teams licensing data to AI labs

Unblock
your data deals.

Close data deals faster. Hand labs a de-identified dataset with a chain of custody for every file, processed entirely in your own cloud account.

Pre-launch · paid pilots, founder-run

Public benchmark · our own runs · Oct 2026

Leak check
0
listed names, emails, phones in 3,736 written files
Held back
479
of 4,215 files held back for review, not written
Recall
96.5%
entity recall on 7 held-out workspaces

Measured on a public 21-workspace de-identification benchmark (v2.5), internal runs. Method →

Works on the files you actually have

  • Email
  • Slack
  • PDF
  • Scans
  • Handwriting
  • Word
  • Excel
  • CSV

And many more

From every system

Wherever your data lives.

Export it. CORPUS Engine de-identifies it in your cloud account.

Databases & warehouses

  • MySQL
  • PostgreSQL
  • SQL Server
  • Oracle Database
  • MongoDB
  • MariaDB
  • Redis
  • SQLite
  • Snowflake
  • BigQuery
  • ClickHouse
  • Databricks
  • Redshift
  • Elasticsearch
  • Cassandra

Team chat & email

  • Slack
  • Google Chat
  • Microsoft Teams
  • Discord
  • Zoom
  • Gmail
  • Outlook
  • Mattermost

Customer & vendor messaging

  • WhatsApp Business
  • Intercom
  • Zendesk
  • Freshdesk
  • Twilio
  • Telegram
  • LINE
  • WeChat

Finance, ERP & payments

  • Tally
  • SAP
  • NetSuite
  • QuickBooks
  • Xero
  • Zoho Books
  • Razorpay
  • Stripe
  • PayPal
  • Adyen
  • Square
  • Paytm
  • PhonePe

Engineering & product analytics

  • GitHub
  • GitLab
  • Bitbucket
  • Jira
  • Linear
  • Confluence
  • Notion
  • Google Analytics
  • Firebase
  • Mixpanel
  • Amplitude
  • Sentry
  • Datadog

Growth, CRM & ads

  • AppsFlyer
  • Branch
  • Adjust
  • CleverTap
  • Braze
  • HubSpot
  • Salesforce
  • Meta Ads
  • Google Ads
  • LinkedIn Ads
  • TikTok Ads
  • Mailchimp
  • Segment

Health records & imaging

  • Epic
  • Oracle Health
  • athenahealth
  • MEDITECH
  • eClinicalWorks
  • NextGen
  • HL7 FHIR
  • HL7 v2
  • DICOM
  • PACS
  • REDCap
  • OpenMRS

CAD, BIM & design

  • Autodesk
  • AutoCAD
  • Revit
  • IFC · BIM
  • DWG
  • STEP
  • Dassault Systèmes
  • SolidWorks
  • Siemens NX
  • Rhino
  • SketchUp
  • Figma
  • Adobe

Gaming & media

  • Unity
  • Unreal Engine
  • Epic Games
  • Steam
  • Roblox
  • PlayStation
  • Xbox
  • Nintendo
  • Twitch
  • YouTube

Files, docs & HR

  • Google Drive
  • Dropbox
  • Box
  • SharePoint
  • OneDrive
  • DocuSign
  • Workday
  • BambooHR

Commerce & retail

  • Shopify
  • WooCommerce
  • Magento
  • BigCommerce
  • Amazon Seller Central

Common sources of the exports CORPUS Engine processes: database dumps, chat and email archives, documents, scans, drawings. Names and logos belong to their owners; no affiliation implied.

How it works

From raw archive to licensable dataset

  1. 01

    Read files

    Text, layout and hidden parts of each file.

  2. 02

    Find entities

    Names, companies, IDs, emails and phones, flagged.

  3. 03

    Link identities

    One identity per person, from your files' evidence.

  4. 04

    Assign personas

    One consistent fake per identity; scans burned out.

  5. 05

    Verify output

    A pass-or-hold verdict per file, with reason.

  6. 06

    Release dataset

    Clean files, review queue, per-file audit report.

Identity graph

Coherent threads. Real identities replaced.

A who's-who map: Erik, E.N. and erik@… are one person, so he gets one fake everywhere and the threads labs buy still add up.

thread-0412.eml Email
Email: names, initials, mailbox, phone and company become consistent fakes.
intake-scan-0193.pdf Scan
Synthetic scanned intake form with a handwritten patient name, date of birth, phone, address, record number, physician and signature. The same form with every identifier, the clinic name, logo and signature burned out; the clinical note is unchanged.
Scanned PDF: identifiers, logo and signature burned out. The clinical note survives.

Synthetic examples. Drag to compare.

  • Erik Nilsson · E.N. · erik@…→ one person
  • R.J.→ not linked: two people match
Re-scan

Nothing ships unchecked.

  • thread-0412.emlWritten
  • intake-scan-0193.pdfWritten · burned out, OCR re-read
  • forecast-q3.xlsxHeld back · name in a hidden sheet
  • slack/general/2026-03-17.jsonWritten
  • board-pack.docxHeld back · name in a comment
Illustrative. Every output is reopened: hidden sheets, comments, footers, properties. A hit holds the file back, with its reason.
Zero data egress

De-identify at the source.

Identifiable data is the risk. Move it, and the risk moves with it. So CORPUS Engine runs where your data already lives, internet disconnected.

  1. Trust

    We never see your data. Nothing to upload.

  2. Control

    Your cloud account, your policy, your keys.

  3. Visibility

    Review every change before anything leaves.

Verified datasetDe-identified, with a per-file report Only exit
Inside your cloud account: raw files, CORPUS Engine, your review Leaves the account: the verified dataset only Internet: no route in or out

The gold standard: identifiable data stays protected until de-identification is complete. Only de-identified data moves.

  • HIPAA de-identification
  • GDPR pseudonymization
  • CCPA
  • Data residency

Built for these workflows. Not a certification.

Your cloud account

Bring your own cloud.

CORPUS Engine runs inside your cloud account on a private network, internet disconnected. You apply our Terraform on a guided call. We never hold credentials and never receive your data. Keep it running, or destroy it.

  • AzurePilots now
  • AWSNext
  • Google CloudNext
  • Red HatRoadmap
  • OpenShiftRoadmap
  • VMwareRoadmap
  • Oracle CloudRoadmap
  • IBM CloudRoadmap
  • Detect
  • Link
  • Replace or redact
  • Custom entities
  • Plain-language policy
  • Review queue
  • Audit report
  • API, CLI & Python SDK
  • Local MCP
  • Your blob storage
Audit report

Every file you sell comes with its paper trail.

Per-file audit report: what was found, what was replaced, what was held back, and why.

Pre-launch · Cohort 1

Prove it on
your own archive.

A paid, fixed-scope pilot on one archive, inside your own cloud account, ending with a per-file audit report. Book a call and we'll scope it together.