The independent verification layer for work performed by AI

AI says it’s done. Vraelis proves it.

Give Vraelis a deployed application and the outcome that should be true. It independently derives what must be checked, verifies the live result, and returns the evidence behind its decision.

Verify an outcome View the API

Starting with deployed web applications.

verification

Runningillustration

From claim to evidence

You describe the outcome. Vraelis works out the rest.

Not a test suite you author and maintain. A sentence about what should be true, turned into checks against the running system.

Describe what should be true

One sentence, about the result rather than the steps to reach it.

“A customer can upgrade to Pro and receive access immediately.”

Vraelis derives what must be checked

It reads the deployed application and works out the requirements that claim implies.

The upgrade action is reachable from the pricing page
Payment completes without an error
The account reflects the Pro plan immediately afterwards
Pro-only capability is actually available
and 2 more

Machine derived. Not human reviewed. Always shown to you.

Vraelis returns a decision, with evidence

Three answers, and the third one is the honest option most tools refuse to have.

Verifiedthe claim held

Failedit did not, and here is what happened instead

Blockedno verdict could be reached

With the step trace, screenshots, console output, failed network requests, and a repair prompt when something breaks.

Independent by design

It checks from outside the application.

Vraelis does not need an SDK, test files, or access to the codebase. It operates the deployed application through a real browser, the way the person receiving it would.

That distance is the point. Anything written inside the system, by the author or the model that built it, inherits the same assumptions the mistake came from. Independence is not something you reach by trying harder. It is structural, and structure is the only way to get it.

AI builder

writes and ships the change

Deployed application

what your customer actually receives

Vraelis checks it independently

from outside, with no access to the source

Failure is not the end

A failed claim becomes something the builder can fix.

A verdict on its own tells you something is wrong and leaves you to find it. The useful output is the package that lets the thing which wrote the code repair it.

Failure observed

Evidence captured

Repair prompt generated

Sent back to the coding agent

New deployment

Reverified

Vraelis packages what should have happened, what happened instead, how to reproduce it, and the browser evidence from the failure.

The coding agent diagnoses the source-level cause. Vraelis independently checks the repair.

Vraelis does not edit your code. It decides whether the outcome is true, and gives whoever wrote it enough to fix it.

Built for people and agents

The same verification, however it is asked for.

For people

Submit and review verifications from the Vraelis application.

Describe the outcome, watch it run, and read the decision with the evidence behind it.

Verify an outcome

For software

Launch and read verifications through the API or CLI.

One command, one exit code. Built so a pipeline or an agent can act on the result without a person reading it.

vraelis verify \
  --url https://example.com \
  --claim "Checkout grants Pro access" \
  --wait

See the developer surface →

Vraelis starts with deployed web applications. The long-term verification layer should work anywhere AI claims an outcome occurred.

That is the direction, not today’s coverage. What is live is on the limitations page, in full.

Before anything says done.

Give Vraelis a deployed application and the outcome that should be true. Get back a decision you can act on, and the evidence behind it.

Verify an outcome See how it works