Privexa · A Translyx Platform
Use powerful AI without exposing more sensitive information than the task requires.
Minimum disclosure. Maximum AI utility.
Text · Documents · Images · Whole-Slide Pathology · Audio · API
Why Privexa exists
Organisations increasingly rely on external AI. Valuable and sensitive information naturally follows the workflow.
Blanket bans rarely hold — they tend to push usage into unmanaged tools. What organisations need is a governed boundary.
Privexa creates that control boundary before AI egress.
Three principles
Information the downstream AI task does not genuinely require does not leave the boundary.
Roles, relationships and context are preserved, so the AI stays genuinely useful.
Your organisation defines privacy and disclosure policy — not the AI provider.
What AI sees
Toggle the policy. Watch what leaves the boundary — and what the authorised user gets back.
Original
Inside your organisation
AI receives
The protected representation that leaves the boundary
Authorised user sees
AI response, reconstructed inside the boundary
Every selected entity type is replaced before AI egress. Placeholders keep roles consistent, so the AI can still reason about the case. Illustrative example with fictional data; no AI model is called.
How Privexa works
Identify sensitive and policy-defined information.
Transform what the policy requires.
Route only the protected representation to an approved model or workflow.
Restore authorised context inside the controlled boundary.
Record the privacy state, model and workflow evidence.
Platform modules
Each module shares the same privacy authority, audit model and organisation policy.
Protected conversational AI with controlled reconstruction.
Work with reports and organisation knowledge without over-disclosure.
Detect, review and protect identifying content in standard images.
Durable, tiled privacy workflows for very large raster images.
Privacy controls for whole-slide pathology — developed capability under validation.
Private, local transcription feeding protected downstream AI workflows.
Organisation controls, audit evidence and evaluation.
Embed the privacy boundary inside your own platform.
Multimodal privacy
Sensitive information doesn't only live in text. Privexa applies the same policy and evidence model across every modality it handles.
Protected AI conversations with controlled reconstruction.
Reports, PDFs, office documents and structured text.
Protected retrieval and organisation knowledge workflows.
Privacy detection, review and protected derivatives.
Durable tiled processing for very large raster images.
Privacy controls for gigapixel pathology slides.
Private, local transcription and protected downstream AI.
Advanced imaging spotlight
Protect identity-bearing slide information while preserving the diagnostic image workflow.
Privexa focuses on the known identity-bearing surfaces of a digital slide — label, macro image, thumbnail and metadata — plus regions a reviewer adds. It does not run generic OCR over tissue, which in testing produced high false-detection rates and harmed diagnostic utility. This capability is developed and currently undergoing extended validation.
Whole-slide image
Gigapixel pyramidal slide, as acquired.
Associated privacy surfaces
Privexa review
Automatic privacy controls · reviewer-added regions
Protected WSI
A new protected derivative, designed to preserve the diagnostic image workflow.
Approval / controlled release
Release is bound to the exact protected version.
Capabilities in depth
Chat with files, extract structure from reports and build organisation knowledge bases. Retrieval runs over protected content, so the model receives the context it needs — not the identities behind it.
Private, local transcription. The transcript is protected before any downstream AI summarisation or structuring, and a human reviews the result.
Organisation administrators set users, roles, approved providers and models. Every AI egress produces evidence of exactly what was authorised to leave.
Provider & model flexibility
Privexa separates organisation privacy policy from model choice, so approved models can evolve without rebuilding the privacy boundary.
Current engineering supports configurable provider and model access. Approved providers and models are set per organisation policy. Trademarks belong to their respective owners; no endorsement is implied.
Deployment
Operated by Translyx. Subject to commercial availability.
Deployed into a customer-controlled cloud environment.
Deployed on customer infrastructure.
For restricted environments, where supported and agreed.
See Privexa in action
Industries
Clinical text, reports, medical imaging and pathology workflows.
Customer information, transaction data, case files and regulated AI use.
Sensitive organisational information and governed AI access.
Documents, datasets and clinical or research workflows.
Organisation-wide AI access with controlled disclosure.
Trust architecture
Privexa can preserve evidence of exactly which protected representation was authorised to leave your organisation's controlled boundary.
Each organisation's data, policy and authority are explicitly scoped.
Approved providers and models are enforced on the server, not chosen by the client.
Protection is derived from versioned state. Reprocessing creates new authority.
Approval is bound to the exact protected representation. Stale approvals do not silently transfer.
Evidence of exactly which protected representation was authorised to leave the boundary.
When protection cannot be established, the workflow stops rather than sending raw data.
Original context is restored only for authorised users, inside the boundary.
A working session with your own scenario — enterprise AI, clinical data, pathology slides or platform integration.