FAQ: agent governance, deployment, and what we do not claim

What it is

Is this an eval tool?: No. Evaluation tells you how a model scores on a benchmark. This states what an agent may do, refuses what exceeds it, and records the refusal. Most teams need both — see how this differs.
Does it slow my agent down?: An authorisation is one call before a consequential action, and the gates are ordered so cheap structural checks run first. An action refused at gate 1 never reaches policy evaluation.
What happens if RotaGrant is unreachable?: Capture fails open and enforcement fails closed: a recording problem must not cause an outage, and an unreachable control plane must not silently permit an ungoverned action. Teams who cannot accept that set a flag and own the consequence explicitly, which is at least a recorded decision rather than an accident.
Does it work with my framework?: It is an SDK call, not a runtime. LangChain, LlamaIndex, your own loop, or an MCP server behind our proxy — the governed unit is the action, not the framework.

What data do you see?: None. It runs in your environment and no payload reaches us on any path. The outbound connections it could make are enumerable and probed rather than described.
Do you call an LLM?: No — not for summaries, not for classification, not anywhere. A test fails the build if a model provider appears in the dependency graph. That is a stronger claim than having an AI feature and a rarer one.
Can it run airgapped?: Yes, and the deployment refuses to start if any setting selects a service outside your network. It also tells you which evidence property you gave up by doing so.

Do you have SOC 2?: Not today — see trust for the position. Most of what a certification would attest to is answered by the deployment model instead: the software runs in your environment and we hold none of your data.
Who are your customers?: We are not naming any. Rotascale has a consulting history in banking, insurance and government, and those engagements are not proof of this product. What we offer instead is a demo running real agents, a published specification, and a red-team grid with our own failures on it.
Does this make my agent safe?: No. It bounds what your agent may do and records what happened. It does not make the agent correct, well designed, or appropriate for the task — and a vendor telling you otherwise is selling you a feeling.
Does it make me compliant?: No. It assembles evidence, maps it to clauses, computes what it can and names what it could not find. Whether that satisfies your obligations is regulatory interpretation, and it rests with you and your counsel — a disclaimer that sits inside the signed payload of every pack, so removing it invalidates the signature.
Why should I believe any of this?: You should not, on our word. That is the point of the marks on these pages: observed links to the thing itself, and asserted says plainly there is nothing here you can check from outside.