On our bare metal.
Allocate dedicated machines or managed workload capacity on our hardware. Integrate provisioning with SystemAPI and policy-controlled adssh execution.
View availability ↗ComputeAPI / Capacity, through code
Your infrastructure.
Your cloud.
Your call.
Completely Programmable Infrastructure starts with capacity. On-demand GPU compute today. Our bare metal, common clouds, and your own provider credentials are next.
Available adapters: Vast.ai + io.net · New capacity models in development
SAH / Architecture01—04
Our infrastructureBare metal / planned
↓
Managed providersVast.ai + io.net / adapters available
↓
Your cloud accountPaid management / planned
Choose the capacity. Program the lifecycle.↗
01 / Three ways to compute
Choose where it runs.
One product direction, with clear choices about hardware ownership, provider accounts, and billing.
Through our providers.
Search GPU offers, provision an instance, inspect it, and destroy it through Vast.ai and io.net adapters. Provider credentials must be configured.
Read the endpoints ↗In your cloud account.
Bring your own cloud credentials. Keep provider billing in your account and pay an additional management fee for our orchestration.
Explore Bring Your Own Cloud ↗02 / An API for capacity
From an offer
to an instance.
The current compute API supports GPU offer search, provisioning, inspection, and destruction. Requests go to your configured ComputeAPI API origin.
Explore the API contract ↗curl "$COMPUTEAPI_URL/api/v2/compute/offers?gpu_type=a100" \
-H "X-API-Key: $COMPUTEAPI_API_KEY"
# Inspect a provider instance
curl "$COMPUTEAPI_URL/api/v2/compute/instances/vast/$INSTANCE_ID" \
-H "X-API-Key: $COMPUTEAPI_API_KEY"03 / Provider coverage
Built for multiple providers.
Available adapters are separate from planned integrations. This is the current implementation boundary.
| Capacity source | Status | Scope |
|---|---|---|
| Vast.ai + io.net | Adapters available | GPU offer search, provision, inspect, and destroy. Requires configured provider credentials. |
| AWS · Azure · Google Cloud · Oracle | Planned | Common-cloud compute adapters and lifecycle operations. |
| Our bare metal | Planned | Allocation, host enrollment, imaging, and physical lifecycle controls. |
| Customer cloud accounts | Planned | Tenant-scoped provider connections with an additional management fee. |
| Managed Docker / k3s capacity | Planned | Connect allocation to SystemAPI workload management. |
The existing Kubernetes namespace API does not yet provision cluster resources. Unified start/stop/restart and automatic SystemAPI enrollment are also planned.
04 / Bring Your Own Cloud · planned
Your account.
Our operating layer.
Use the cloud relationships you already have. The planned Bring Your Own Cloud offering adds orchestration and lifecycle management while the infrastructure stays in your provider account.
Two bills. Clear ownership.
You pay your cloud provider directly for resources. ComputeAPI charges an additional management fee. Pricing will be published with availability.
Connections scoped to the customer.
The design calls for customer-specific provider connections using limited credentials or delegated roles, with policy and audit across operations.
Discuss your cloud requirements ↗Capacity is only the beginning.
SystemAPI and the SAH tools provide the management and execution foundation for the upcoming lifecycle.