PROTECTION
Know what the model received
Compare the original request, protected payload and restored response. Check supported tool calls as well as chat.
AI redaction gateway · Australia
Give AI the context.
Keep control of the identities.
Names, contact details and account references don’t belong in every model request. Veil replaces them with consistent placeholders, then restores them in the answer.
A$490 ex GST guided pilot · No per-seat gateway fee · Built around your workflow
YOUR APPLICATION
Help Alex Morgan
with a $49.95 refund.
Veil
Identity map retainedYOUR RESPONSE MODEL
Help [[PERSON_01]]
with a $49.95 refund.
Keep the useful fact. Replace the identifier.
Illustrative example, not a live model response.
Consistent identitiesVisible rule learningHosted or local detector assessmentYour response-model choice
See exactly what crosses the boundary.
Follow one request from the application to the model and back. Change the workflow, step through it yourself, or play the walkthrough.
The useful part of the request should survive.
A model can draft a refund response without knowing the customer’s real name.
But it still needs the amount, the issue and the next step. Veil is built to protect identifiers while preserving that working context.
Follow the complete workflow →Protect
Person names, email addresses, phone numbers, postal addresses and account-style identifiers.
Preserve
The business facts needed for the task: amounts, products, deadlines and instructions.
Verify
Both sides of the trade-off. A missed identity and a lost business fact are different failures.
A gateway you can inspect.
REUSE
Watch rules become useful
Exact-value mappings and validated field rules reduce repeated work. Changed free text still needs detection.
EVIDENCE
Make the decision with numbers
Review protection, task preservation, detection cost and first-use versus repeat latency.
One boundary. Different responsibilities.
01020304
Business leaders
Start with one useful workflow. See what gets protected, what the response model receives, and whether the result is worth the operating cost.
CIOs & IT leaders
Put a redaction gateway between an application and its response model. Keep detection, deployment and model choice separate.
CISOs & privacy teams
Check which identifiers were replaced, what remained, and who can reveal the originals. Make exceptions and limitations visible.
Application teams
Protect chat messages, structured context and supported tool calls, then restore the identities your application needs.
Choose where detection happens.
Start with a supported workflow.
Evaluate the gateway on Workers AI, with the dashboard, playground, rules and reports. Original input is processed by the hosted detector.
Understand the hosted boundary →Keep detection on your infrastructure.
A tested local API can run the detector and mappings on a CPU VM. Your response model can remain elsewhere. The full dashboard is currently hosted.
See measured CPU results →A few important questions.
What does Veil protect?
The current detector covers person names, email addresses, phone numbers, street/postal addresses and personal account-style identifiers in supported text and structured fields. It is not a general file, endpoint or email security suite.
Will the model still understand the request?
Veil replaces identifiers with consistent placeholders and leaves useful business context in place. The response is restored afterwards. Evaluate your own examples: a detector can miss an identifier or remove something useful.
Can original personal information stay on our infrastructure?
The local CPU API proof of concept runs detection and mapping storage locally, while the response endpoint can be hosted elsewhere. The public hosted pilot processes original input on Workers AI. The full dashboard has not been ported into an on-premises product.
Does it work with our existing model?
Model choice is separate from redaction. The hosted gateway currently uses configurable Workers AI response models; the local API can use an OpenAI-compatible response endpoint. Provider behaviour, streaming and tool calls are checked during your pilot.
Start with evidence
One workflow. A clear decision.
Bring the use case. We’ll help you test what gets protected, what stays useful and what it costs.