How it works
No test scripts to write and no dashboard to babysit. You define what must work and approve what runs; Vraelis executes it against the exact build, captures the evidence, and returns one truthful verification decision, with the reasons behind it.
The loop
Give Vraelis your deployed URL and the prompt you built the app from. Connect GitHub, Vercel, Supabase, or Stripe to give the plan setup and context; the evidence comes from what the real browser observes.
Vraelis reads the app and the prompt to learn what it promises, then links each promise to the layer that has to keep it: a route, a database write, an ownership rule, a charge.
You get an editable contract of what the app must do. Nothing runs until you approve the critical flows, so Vraelis tests only what you signed off on.
A real isolated browser runs your approved flows as two separate users. Every step is a deterministic observation of what was clicked, what rendered, and what the network and console did. No model guesses here.
One decision: Verified, Failed, or Blocked. Each failure carries its requirement, the expected and observed behavior, exact reproduction steps, and a screenshot.
Every failure ships with a fix prompt for your builder. Push the fix and Vraelis reruns the exact failed check on a preview, then confirms the regression is closed.
What running it like production means
A green screen is easy. Vraelis asks the harder questions: did the data actually get written, does it survive a refresh and a new session, and can a second user reach it. These are first-class issue types, not footnotes.
How Vraelis stays honest
The pass or fail of every flow comes from what the browser actually did, never from a model's opinion. Screenshots, step timelines, and console and network activity are the record. When Vraelis suggests a likely cause, it is marked as interpretation and never counts as evidence.
The decision follows one explainable rule, not a number you have to trust: any critical flow that fails means the claim comes back Failed, anything that needs a human call comes back Blocked, and only a clean run is Verified.
The output
Every run ends in one answer you can act on, each with the reason and the evidence behind it.
Connect your AI-built app and get a decision before your users find the failures.