NEW · REPLAY LIVE: A CISO's Guide to Proving Agentic AI Governance

Watch

Trinitite

PricingResearchBlog

Reversible Masking

Use the best AI. Keep your secrets.

The best AI lives outside your walls. But your customer data, health records, and account numbers can't go there in the clear — not without failing an audit.

We hide the secrets before they leave, keep your workflow running, and put the real values back only at the doors you trust.

The Bad Trade

Every other option makes you give something up.

Black out the data

Plain "[REDACTED]" strips the value entirely. The AI loses the thread, and you get back answers full of blanks to patch by hand.

Trust the provider

Hand the raw data over and hope. That is a position that cannot survive a HIPAA audit, a GDPR review, or a bank exam.

Run a weaker private model

Keep the data in-house, but trade away the smartest models on the market. You solve the leak by giving up the capability.

Reversible masking is the option that doesn't make you choose. The model gets enough to be useful. The secret never leaves in the clear.

How It Works

Hide it. Work on it. Put it back.

You change nothing in your app. Trinitite already sits in the path of your AI traffic. Masking is two quiet hooks in the path you already use.

Reversible Masking — Live

claims-handler flow

Your app

AI provider

sees tokens only

Trusted edge

Original request (inside your walls)

Patient Jane Doe, SSN 482-91-3847, account 4821-7739 — summarize the overdue balance.

Sensitive values detected before anything leaves your network.

vault: in-memory · shred: 1 call

01

We spot the secret

As a request leaves your network, we catch the sensitive bits — names, Social Security numbers, account numbers, health details, your own custom fields.

02

We swap in a stand-in

Each secret is replaced with a consistent stand-in token. The AI sees a complete, sensible request it can actually work with — just not the real values.

03

The work gets done

The AI and your tools reason over the stand-ins. The same secret reads as the same token across every step, so multi-step agents never break.

04

We restore at the trusted exit

On the way back, the real values reappear — but only at the exits you have explicitly trusted. Everywhere else, the secret stays a token.

Zero

App Changes

In Memory

Never The Database

1 Call

To Erase A Job

Allow-List

Default Deny

What You Walk Away With

Full AI power, with the secrets locked down.

Use the AI you actually want

No more choosing between a capable model and a private one. Send your work to the best model on the market — it just never sees the secret parts in the clear.

Multi-step agents keep working

Plain "[REDACTED]" blanks out the value and breaks the agent’s train of thought. Our stand-ins keep the workflow whole from the first step to the last.

Default-deny on the way out

A tool only sees a real value if you put it on the trusted list. Everything else gets a token. Trusting a new tool is a deliberate choice, never an accident.

One button erases it all

The real values live in memory, never in your audit database. A single call shreds a whole job’s secrets for good — perfect for "delete my data" requests and incident response.

Built For The Auditors

Proof you masked it — without ever showing the secret.

Every job leaves a signed record of exactly which kinds of data were hidden — so an auditor can confirm the control was working without ever seeing a single real value.

HIPAA

Patient details are hidden from the model and restored only at the points your covered entity has trusted.

GDPR

Pseudonymization by design, plus a one-call erase for right-to-be-forgotten requests.

GLBA

Customer financial data is masked before it reaches any outside model, with per-tool allow-listing.

Stop choosing between smart AI and safe data.

See it run on one of your real workflows. Watch the secrets turn to tokens on the way out and come back whole at the door you trust.