Runtime supervision
Host and expose privately operated model runtimes through a consistent service boundary.
Private model infrastructure
PLM hubs and nodes.
The infrastructure layer of GIC: operator-hosted model execution, device identities, network boundaries, and verifiable connections.
PLMH · Execution layer
A private hub between model runtimes and applications.
PLMH is GIC’s hub architecture for operating private models behind a controlled protocol boundary. Its job is to make model discovery, authentication, health, capabilities, and streaming explicit.
Model registry
Represent providers and models through canonical identities and explicit capability discovery.
Authenticated access
Pair devices and authorize model operations rather than treating network reachability as authorization.
Provider adapters
Connect local model engines through adapters while keeping the hub protocol stable for clients.
PLMN · Access & verification layer
A private AI compute fabric across devices, networks, and infrastructure.
PLM is designed to extend beyond a model running on one machine. The direction is an operator-controlled fabric where models can execute on-device, on an in-house workstation or GPU server, across an authorized home, lab or office network, or in private cloud infrastructure when additional capacity is required.
The architectural objective is simple: models, compute, network access and data remain under an explicit operator-controlled trust boundary rather than being implicitly delegated to a third-party model API.
1CONFIGURED
2DISCOVERED
3REACHABLE
4PROTOCOL_OK
5AUTHENTICATED
6MODEL_READY
7STREAM_READY
Deployment direction
One private model layer, multiple places to compute.
On-device
Phones, laptops and edge devices can run capable local models where hardware permits.
On-premises
Workstations and GPU servers can provide larger private model runtimes to authorized devices on a home, lab or office network.
Private network
PLM can discover and route across authenticated nodes while keeping compute inside the operator-controlled trust boundary.
Private cloud
Customer-controlled cloud or VPC compute can extend capacity without making a third-party model API the default execution path.
Hybrid routing
A future PLM fabric can select local, on-premises or private-cloud compute according to capability, policy and availability.
Your models. Your compute. Your network. Your data.
This is the PLM roadmap and architectural direction; deployment capabilities vary by component and maturity.