CogniveilDemo

Regulated work, with the proof attached

AI for work that gets checked.

Cogniveil makes regulated workflows deployable on AI, producing the work and the proof together.

Cogniveil

Audit control hub

Governed run
IA

ICT Contract Audit Coworker

DORA supplier portfolio · Reports to Head of Internal Audit

Phase 01 · Mandate

Scope the ICT supplier audit

Scope locked

The coworker locks the population, applicable requirements, materiality rules, and accountable owner before reviewing evidence.

ControlEvidenceOutcome
Supplier population42 approved ICT agreementsIn scope
Regulatory standardDORA Articles 28 and 30Applied
Audit mandateTPRM-2026-04 · approvedVerified

No evidence can enter from outside the approved mandate.

Population, rules, and ownership are sealed in the run record.

ISO 9001

ISO 27001

ISO 27701

ISO 22301

DORA·Compliant by design, with an ongoing program

EU AI Act·Mapped to

GDPR·Built to support

NIS2·Aligned with

AI can produce the answer. You still cannot use it.

A general-purpose model has no permissions, no accountability, and no trace. It answers from training data, cannot show where a claim came from, and leaves no record you could put in front of a regulator. In a regulated function, that is disqualifying, however good the answer sounds.

There is a cost problem too. AI produces the first version. A qualified person checks it. If that person has to reproduce the work to trust it, the benefit disappears. Cogniveil turns verification from doing the work again into checking the work.

What the platform enforces

Model first

Policy and proof are added after the answer.

01 · Policy overlayA rule around the model, not an enforced gate.
02 · Model firstTraining data and a prompt produce the answer.
03 · Human verificationA qualified person repeats the work to trust it.
04 · Record assembled laterEvidence is reconstructed after the decision.
CogniveilHyperscalers

Hyperscaler AI puts the model first and adds policy and review around the answer. Cogniveil enforces identity, approved sources, verification, and approval before retaining the complete record.

The hyperscaler stack puts the model first and policy on top. Cogniveil inverts it: controls come before the model, so nothing reaches the model, or leaves it, outside them.

How it works

Three things hold on every piece of work, and they are the whole product:

Show your work. You can show exactly how it was done, years later.

Check the source. Every sentence and number points back to where it came from.

Stays where you put it. Your information never leaves the place you keep it.

Underneath, the work is done by AI coworkers: named roles with a job description, approved sources, guardrails, and a human supervisor. The coworker does the work. The human keeps the decision.

Watch one piece of work go end to end: request, work produced, sources, calculation, approval, immutable record, and final artifact.

ICT supplier audit workflow

Locking the approved audit mandate

Follow a sample answer from question to approved source, then open the cited document at the exact supporting passage.

Cogniveil
Enterprise Index
The demo is interactive. Open Workflows or Chat.

Cowork

Your teams

Pick a business team to see its coworkers, chat one-on-one, or start a business briefing.

TeamDescriptionMembers

Select Third-Party Risk Team to start the demo.

See it before you talk to us

A redacted output with the full citation chain, from a synthetic environment. No client data appears in it.

View the sample here, no email needed.

Who it is for

Start with the function that owns the judgment.

Banking & Finance. Remove the second pass.

Close evidence, reconciliation, and journal entry testing are produced with the proof attached, so review means checking, not redoing.

Risk and compliance. Reconstruct the decision later.

Show who asked, what was retrieved, what was answered, and who signed when a regulator asks how a decision was made.

Audit. The workpaper is the product.

Evidence extraction, sampling, and reconciliation are computed deterministically, with a defensible record of every decision.

Legal. The lawyer keeps the judgment.

Contract review and comparison keep every finding mapped to the law it comes from. This is not legal advice.

All solutions

When the evidence is not there

What it does when it does not know.

Most AI fills the gap with a plausible guess. When Cogniveil cannot ground a claim in the approved sources, it says so instead of guessing, and that response is itself recorded.

Verification and citations

Regulatory assistant

Approved sources only

Ready

Checking approved sources

0 of 3
Policies
Supplier contracts
Regulatory library

Preparing grounded response

Trust

Built from the controls a regulated buyer has to prove.

We do not train on your data. Access is enforced at the identity layer, every action enters a tamper-evident audit trail, human approval sits where it belongs, and the platform can run in your cloud, on your premises, or fully air gapped.

Evidence for your review

Get the security pack

For security reviewers and vendor risk teams. The pack contains the four ISO certificates with their full scope statements, the architecture diagrams, and the data processing agreement. The Statement of Applicability is included in the pack; it is not published on the site.

Submitted details are used to respond to this request.

Deployment without compromise

On premises when it must be. Hybrid when it should be.

Put Cogniveil inside your environment, or place each part of the workload across an agreed hybrid boundary. The infrastructure changes. The permissions, citations, approvals, and audit record do not.

Explore deployment optionsAir-gapped operation supported

Controlled boundary

Same governed platform

On premises

Inside your environment

Run Cogniveil on customer hardware or as a managed appliance within the network boundary you control.

Hybrid

Across an agreed boundary

Place governed data and services where policy requires, with only approved connectivity crossing the boundary.

Data placement agreed
Private connectivity
One evidence chain

How it is priced

Priced per coworker, not per seat.

You pay for a unit of work that has a job description and measures, not for how many people log in. That keeps the cost tied to what actually gets done.

See the pricing model

Start with one workflow

Bring the workflow. Define success. Run it. Measure it. Keep the evidence.

Bring us a workflow

Contact sales@cogniveil.ai