The confidential agent layer for sovereign AI clouds.

Terrabox is the software layer that makes AI agents operator-excluded, attested, governed, and audited — running on your GPU fleet, in your data centre, under your sovereignty. We bring the trust layer; you keep the infrastructure and the customer.

For AI cloud providers View SDK Interface
ROOT_OF_TRUST INTEL_TDX_ATTESTED
RUNS_ON YOUR_SOVEREIGN_CLOUD
EGRESS DEFAULT_DENY_GOVERNED
AGENT_IDENTITY ED25519_+_VAULT

The trust layer your agents run inside — wherever they run.

Terrabox turns any confidential VM on your fleet into a verifiable control surface: kernel lockdown, attestation state, audit-chain status, tenant isolation, and credential revocation. Your GPUs, your data centre — our attested runtime on top.

terrabox-a3-tdx-h100 UTC 02:59:01
KERNEL_LOCKDOWN CONFIDENTIALITY (ENFORCED)
ATTESTATION INTEL TDX VERIFIED
AUDIT_LOG HASH_CHAIN_VALID
API_AUTH ARGON2ID_MTLS_READY
KILL_SWITCH AGENT_REVOCATION_TESTED
$ sudo ./scripts/enforce_kernel_lockdown.sh --require

For sovereign AI clouds & neoclouds.

Offer your tenants confidential agents without building it. Terrabox drops in as an ISV layer on your platform — complementary to your infrastructure, never competing with it.

complementary

We don't sell GPUs

We make the agents running on yours confidential, governed, and auditable. Zero overlap with your compute, cooling, or model marketplace — we sit a layer above.

You provide
GPUs, data centre, sovereignty
We provide
The attested trust layer
one_click_isv

One click for your tenants

List Terrabox as a one-click add-on in your console. A tenant who needs confidential agents enables it themselves — you don't build, staff, or maintain it.

Integration
SDK + API, Kubernetes-native
Runs
Inside your confidential VMs
you_keep_the_customer

Your brand, your billing

White-labellable. Your customer relationship, your invoice, your data-residency guarantee. Nothing leaves your fleet; we never see tenant data.

Data
Stays on your infrastructure
Evidence
Tamper-evident audit per tenant

Three deployable agent templates.

Built for regulated workflows where identity, auditability, and data isolation are non-negotiable.

fraud_detector

Autonomous Fraud Detection

Financial agent enclaves with hardware-bound transaction limits.

Control
Budget and API-call guardrails
Evidence
Signed execution trail
demand_forecaster

Sovereign Demand Forecasting

Supply chain modeling isolated entirely from platform operator visibility.

Control
Tenant-encrypted data sources
Evidence
Operator plaintext access denied
compliance_monitor

Continuous Compliance Monitoring

Real-time hash-chained signed audit trails for automated governance.

Control
Hash-chained audit logger
Evidence
Daily integrity verification

A production path a developer team can use in an afternoon.

terrabox-sdk 1.0.0 exposes deployment, execution, revocation, dashboard retrieval, mTLS configuration, and client-side encrypted data-source registration.

pip package: terrabox-sdk==1.0.0 auth: bearer API key + mTLS tenant data: AES-256-GCM before upload 
from uuid import UUID
from terrabox_sdk import TerraboxClient

client = TerraboxClient(
    "https://api.terrabox.example",
    api_key="tbx_live_keyid_secret",
    require_https=True,
    client_cert_path="/secure/tenant/client.pem",
    client_key_path="/secure/tenant/client-key.pem",
    ca_cert_path="/secure/tenant/ca.pem",
)

tenant_id = UUID("00000000-0000-0000-0000-000000000001")

agent = client.deploy_agent(
    tenant_id=tenant_id,
    name="fraud-prod",
    template_id="fraud_detector",
)

result = client.execute_agent(
    agent["agent_id"],
    {
        "transaction_id": "txn_001",
        "amount": 1250.0,
        "merchant_country": "AU",
        "device_seen_before": False,
    },
)

dashboard = client.agent_dashboard(agent["agent_id"])
client.revoke_agent(agent["agent_id"], reason="rotation")