Operational GuideOperating Attexa Sovereign Agents
Sovereign Agents run on infrastructure you control. This guide shows exactly where that can be: which clouds, which instances, and what it takes to enable each, so you can see your deployment options at a glance.
If it can run a hypervisor, it can run Sovereign Agents.
Every agent runs in its own hardware-isolated microVM. So the platform's only hard requirement is access to hardware virtualization (KVM) on the host.
That single rule decides everything below. Any compute you can run a hypervisor on qualifies, which today includes the major clouds, dedicated bare metal, and single-tenant hardware. The capability is now broadly available across providers; the difference between them is mostly which instance families expose it and how you switch it on.
"Can this compute run a virtual machine?" If yes, Sovereign Agents can run there, with the full governance and evidence layer intact. If the only thing on offer is fully-managed software with no host access, it cannot, because there is no virtualization layer to use.
Supported infrastructure, by provider.
Find your provider, confirm an eligible tier, and note how virtualization is enabled. Custody character is tagged on the right: In your cloud runs inside contracts they already hold; is single-tenant hardware; Zero egress is the closed-loop option.
| Provider | Eligible tier | How virtualization is enabled |
|---|---|---|
| Major clouds: inside contracts you already hold In your cloud | ||
| AWS | EC2 8th-gen Intel (C8i, M8i, R8i, X8i families and flex/d variants) | Launch with nested virtualization enabled, or set it on a stopped instance |
| AWS | EC2 7th-gen Intel (C7i, M7i, R7i, I7i and flex variants) | Same nested-virtualization flag (available June 2026) |
| AWS | EC2 bare metal (any .metal type) | Direct KVM access by default, no flag |
| Google Cloud | Compute Engine Intel (N1, N2, C2, C3, C4, M1/M2/M3) | Create the instance with nested virtualization enabled |
| Azure | Virtual Machines (Dv3/v4/v5, Ev3/v4/v5, Fsv2, select AMD v5/v6) | Enabled by default on supported SKUs; choose an eligible series |
| Cost-efficient & bare metal | ||
| DigitalOcean | Standard Droplets with VMX present | Load the KVM module after provisioning |
| Hetzner | Cloud Intel plans, or dedicated servers | KVM exposed by default on most Intel plans; verify on the instance |
| Equinix Metal, OVHcloud, other bare metal | Dedicated hardware | Direct KVM access, no configuration |
| Strictest custody | ||
| Dedicated datacenter | Owned or leased, single-tenant hardware | Direct KVM access; full physical custody |
| Closed-loop enclave | Hardware-attested confidential computing Zero egress | For workloads requiring zero data egress |
Match the requirement to the tier.
These can be mixed. A common pattern is dedicated hardware or an enclave for the most sensitive workflows and their existing cloud for everything else, with the same governance and the same evidence model across both.
The fastest way to start is one question: which cloud and which region does your compliance program already cover?
From there, we confirm the eligible instance types, deploy into the environment you choose, and your team verifies the first actions from a witness node you control. Wherever it lands, the governed action boundary and the customer-held evidence are identical.