rotascale authority and evidence for AI agents See the demo

Deployment

It runs where your data already is.

Single-tenant, in your environment. Not a tenant in ours with a residency label — no payload reaches us on any path, and the outbound connections a deployment could make are enumerable and probed rather than described.

Four tiers, and what each one gives up

TierRunsSigning keyWhat it forbids
Sandbox Ours, synthetic data only Ephemeral — and every artefact says so Real data. It is for learning the shape.
Your cloud Your VPC, your account Your KMS or HSM Nothing by default; egress is per-feature and off unless set
On-premise Your data centre Yours, up to external custody Anything your network does not permit
Airgapped No route out at all Yours Every outbound path — including external anchoring

Airgapped costs you a property, and the platform says which. External anchoring is what makes integrity independent of trusting the operator. Forbid the path and that independence is gone — so the deployment states it rather than quietly dropping the mechanism and leaving the claim on the page.

Custody decides what a signature is worth

deployment
The key is on the server. The operator could re-sign a rewritten history, and an auditor is entitled to say so.
kms / hsm
Non-exportable. They can sign new records; they cannot extract the key.
external
You hold it. RotaGrant can ask for a signature and cannot produce one alone — withdraw access and sealing stops, which is a check you can run rather than a promise you have to take.

A private key is never uploaded, pasted, downloaded or exported. The capability is absent from the interface rather than present and forbidden.