Sovereign Offline Orchestrator for Department-Partitioned Artificial Intelligence Brain-Fleet Inference on a Single Offline Computing Device.

Partitions the brain fleet into department-scoped enclaves on a single offline device, provable from a public key alone. 36 claims.

An orchestrator on a single sovereign offline device that partitions the cooperative artificial-intelligence brain runtime into department-scoped fleets, each fleet defined by a fleet-scoped post-quantum keypair fuse-bound to the device hardware root-of-trust, a whitelisted subset of brains, a separate append-only audit ledger sealed under the fleet keypair, a hardware-attested memory partition and a signed knowledge-base, with cross-fleet aggregation requiring an explicit cross-fleet quorum action sealed to a top-level ledger. The partitioning is provable from a public key alone, with no network in the trust path. Defended against cloud-hypervisor multi-tenant isolation (AWS IAM, Azure RBAC, K8s namespaces) which depends on a network authority not present in a sovereign offline device, against single-tenant inference frameworks (single-tenant inference servers such as vLLM, Triton and TensorRT) which provide no isolation at all, and against confidential-computing approaches (Intel TDX, AMD SEV-SNP, ARM CCA) which protect a single tenant from the host but do not partition multiple tenants on a single offline device under post-quantum bindings to a hardware root-of-trust. Filed 25 June 2026 as GB2615041.7.