When client documents can't go into ChatGPT, Claude or Copilot, useful AI and client confidentiality hit a deadlock.
ZTHI breaks it, on your machine: it finds potential sensitive values and puts every detection in front of your reviewer.
Only the reviewed, tokenised text goes to the AI you choose. The real details stay local — and the session leaves a tamper-evident record.
The synthetic specimen below is loaded and ready — or paste text of your own. Detect, review each finding, swap, see what the AI would receive, then restore. Exactly the workflow your team would use — nothing here leaves your machine.
Honest scope note: this page uses simple pattern matching and a scripted reply — a taste of the workflow. The desktop engine adds AI-based entity detection, more categories, PDF ingress, and the sealed evidence register you can hand to an auditor. Nothing you paste here is transmitted, stored or logged; this page has no server side.
What was protected, who approved it, when — sealed with a tamper-evident chain and a digital signature from the installation. The certificate contains fingerprints only — designed to be shareable with a regulator, client or board without exposing the protected values.
Every entry links to the one before it. Alter anything and the chain — and the signature — stops verifying.
The record carries what categories were protected and that approval happened — never the values themselves.
The machine-verifiable proof travels with the certificate and re-verifies on your own machine, any time.
ZTHI — REDACTION CERTIFICATE (safe to share — contains no personal data) Processed: 2026-07-18 12:46:21 Reviewed & approved by: reviewer (on this machine) Sealed: 2026-07-18 12:51:41 UTC WHAT WAS PROTECTED (136 items) Address................. 23 Date.................... 12 Company/Entity.......... 19 Passport................ 8 ID Number............... 17 Date of Birth........... 7 Person.................. 16 Email................... 7 Amount.................. 13 (and 5 more categories) INTEGRITY Tamper-evident record: INTACT (280 linked entries) Digital signature: VALID (genuine, from this installation) Personal data herein: none — fingerprints only
On your machine, the engine finds potential sensitive values — names, addresses, account numbers, identifiers — across PDFs and pasted text.
Each detection goes in front of your reviewer, who approves, removes or adds items. Detected is the machine's job; approved is a person's.
Approved values are replaced with neutral tokens. The keysheet that maps them back is private and stays on the machine — deliberately separate from anything shareable.
You copy the reviewed, tokenised text into whichever AI your firm chooses — ZTHI never touches your AI accounts — and paste the reply back.
The real details are restored locally, on your machine. The session seals its tamper-evident record: what was protected, who approved it.
One subscription, priced for the Island. Quarter-to-quarter — leave at any quarter end. First invoice at pilot acceptance, not at install; we set everything up with you.
| Component | Per quarter |
|---|---|
| ZTHI desktop | Unlimited use on supported inputs (PDF and pasted text), subject to mandatory review, on your machines. Supervised install — we set it up with you. |
| IoM Regulatory Radar | Monthly issues: what moved in Isle of Man financial-services regulation, cited to the regulator's own pages. |
| Horizon scanning | A quarterly forward-look briefing — what's coming, what to prepare — plus alerts when new statutory documents touch your obligations. |
| Screening support | Up to 50 sanctions-screening checks per quarter against our consolidated sanctions list store. Turnaround terms in your agreement. |
| Staff training | Safe AI use in a regulated firm — quarterly module scope confirmed before your first invoice. |
No setup fee · cancel at quarter end · ZTHI's design rule applies to everything we ship: detected is the machine's job, approved is a person's — outputs must be verified by a qualified professional.