A flight simulator isn’t cheaper because it doesn’t fly.
The most common question about Control Studio isn’t a technical one. It is: „Why should I pay for software that doesn’t actually execute anything?“
That is a fair question, and it deserves a better answer than a feature list. A flight simulator isn’t cheaper because it doesn’t fly. It is valuable because it doesn’t crash. The value lies not in the effect, but in being able to play the effect through beforehand — with every consequence, and without any wreckage.
With AI agents the situation is exactly the same, except that almost nobody talks about it. A language model can analyse an invoice, propose a payment and phrase the reasoning behind it — fluently, convincingly, in seconds. What it cannot do is guarantee that this payment complies with a binding policy, that a second person approved it, and that the case remains auditable afterwards. That gap does not disappear with the next, larger model. It is structural.
The mistake I run into most often is treating governance as something you write down. A policy document gets drafted, reviewed, signed off — and then it goes live without anyone ever having checked whether it can actually be executed. Whether the rules contradict each other. Whether an approval fires at the right point in the flow. Whether the case you described really gets blocked when it matters, or only almost. That is roughly like certifying an aircraft on the strength of its manual.
ZERYON Control Studio is the workbench for the step before that. Agents, workflows and policies are built here, tried on real cases and made provable — before an AI is allowed to act in production. The flow is always the same and always visible: agent, workflow, policy check, approval, execution, audit ledger. The agent keeps its analytical strength. It only loses the authority to execute its own proposals unchecked.
Control Studio is deliberately not a chat. That isn’t a convenience decision; it is the heart of the matter. In a chat it blurs who decided. Here the control decision is made in rule-based, reproducible code — the same input under the same policies yields the same decision. That is precisely what makes it possible for an auditor to retrace the decision later.
Three mechanisms separate a described governance from a proven one. First, precedence in the policy engine: when several rules match, deny beats require_approval beats allow, and a deny creates no approval request at all. A block cannot be lifted by any authorisation — which sounds unspectacular but is exactly where most home-built approval logic gives way. Second, separation of duties: whoever starts a case cannot approve it, and whoever authored the triggering policy cannot approve the high-risk action it fires. Technically enforced, not recommended in a procedure. Third, the ledger: append-only at database level, cryptographically chained, with the event hash keyed by a secret held outside the database in the operating system’s key store. Write access to the file alone is not enough to recompute the chain. The rejected attempt is recorded too.
And because execution runs in simulation, you see the full governance effect without risking any real side effects. No payment, no email, no record written. Only the question of whether the control system does what you claimed it does.
It is worth being clear about what Control Studio is not. It is not a runtime control layer and no substitute for production. Once the governance holds and real effects are to be connected, AXYON takes over enforcement. Control Studio is the place before that — design and evidence, not execution. Confusing the two means buying the wrong tool.
What you hold at the end of a day in the Studio is not a concept paper but a verified evidence export that audit and compliance can read. The difference between „we have an AI policy“ and „we can show that it holds“ is exactly that export.
Call to action: We walk through a real case: design a policy, start a case, force an approval, verify the evidence — in under an hour. After that you’ll know whether your AI governance is executable or merely written.
Frequently asked questions
Why pay for software that doesn’t execute anything for real? Because the value is in the rehearsal, not the effect. Control Studio costs a fraction of what a single unchecked AI action in a production system can cost — and the single-seat licence is €1,900 one-off, net plus VAT.
Are real payments or emails triggered? No. Actions run in simulation. You see the full governance effect — check, block, approval, ledger entry — without any real side effects. Once real effects are to be connected, AXYON takes over.
Can an approval be bypassed? No. A deny creates no approval request at all and cannot be lifted by any authorisation. On top of that, separation of duties is technically enforced: no self-approval, and no approval by the author of the triggering policy. The rejected attempt is recorded in the ledger as well.
Is a single-seat licence enough to try it out? For designing, yes. For demonstrating, no. Separation of duties is technically enforced — whoever starts a case cannot approve it. A single workstation therefore cannot show the core function; that is what the team licence is for.
Does the data stay in-house? Yes. Control Studio runs locally as a native macOS application with its own data store and needs no internet connection for core operation. Minimal system entitlements, no shell execution, no unrestricted file access; shipped signed and notarised by Apple.