How on-premises AI inference servers operate fully air-gapped for SCIF environments, classified networks, remote field operations, and network-isolated facilities.
Yes. On-premises AI inference servers operate entirely on local hardware with no internet dependency. Models are pre-loaded, inference runs on local NVIDIA GPUs, and the system functions identically whether connected or fully air-gapped. This is critical for SCIF environments, classified networks, remote field operations, and any facility where network isolation is a security requirement.
An air-gapped AI system has no physical or logical connection to the internet or any external network. The server operates on its own hardware, with all software, models, and configuration pre-loaded before deployment. No data enters or exits the system via network connection. Users access the AI through a local area network or direct physical connection within the secured facility.
This is not a software toggle that can be accidentally reversed. Island Mountain pre-configures the system for air-gapped operation by disabling all external connectivity features at the environment variable level before the server ships. Your IT team can verify each setting independently and confirm that no outbound connections are possible.
The following OpenWebUI environment variables are configured for air-gapped operation on every Island Mountain system:
OFFLINE_MODE=true - Disables all outbound network requests from the application layer.
HF_HUB_OFFLINE=1 - Prevents Hugging Face model hub downloads. All models are loaded from local storage only.
ENABLE_COMMUNITY_SHARING=false - Disables community prompt sharing that would require external connectivity.
ANONYMIZED_TELEMETRY=false - Disables all anonymized usage telemetry that would phone home to external servers.
ENABLE_RAG_WEB_SEARCH=false - Disables web-based retrieval augmented generation that would require internet access.
SAFE_MODE=true - Enables restricted mode preventing execution of potentially unsafe operations.
ENABLE_SIGNUP=false - Disables self-registration. Only pre-configured administrator accounts can create new users.
Full technical documentation is available on the technology page.
Defense contractors: SCIF environments and classified networks governed by ICD 503 and CNSS Instruction 1253 require physical network isolation. On-premises AI brings inference capability inside the security perimeter without breaching the air gap. Learn more about defense contractor AI infrastructure.
Energy and utilities: NERC CIP-protected control system environments require network segmentation. AI inference on isolated operational technology (OT) networks supports grid analytics without exposing critical infrastructure to internet-connected AI services. Learn more about energy sector AI infrastructure.
Government agencies: Federal and state agencies handling CUI under NIST SP 800-171 benefit from air-gapped AI that stays within their existing security authorization boundary. No FedRAMP dependency, no cloud provider contract, no external data flow. Learn more about government AI infrastructure.
Remote and field operations: Tribal lands, rural clinics, disaster response forward operating bases, research stations, and maritime vessels - anywhere internet connectivity is unreliable or nonexistent. The system processes queries on local hardware with zero internet dependency.
Model updates are performed via physical media transfer. New model files are downloaded to a separate internet-connected workstation, transferred to approved removable media (USB drive, external SSD), scanned per your organization's security protocols, and loaded onto the air-gapped system. This process ensures the air gap is never breached and all updates pass through your security review before deployment.
Model updates are optional - the system runs the same models indefinitely without updates. Many air-gapped deployments intentionally freeze their model configuration to maintain a known, tested baseline. Updates are available when your team decides they are needed, not when a vendor pushes them.
An air-gapped AI system has no physical or logical connection to any external network. All software, models, and data stay on the local hardware. Users access the system through a LAN or direct connection within the secured facility. No data can enter or exit via network.
No. The inference engine runs identically regardless of network connectivity. The same models, the same GPU hardware, the same response quality. The only difference is that web search augmentation and community sharing features are unavailable - which is the point of air-gapping.
Yes. Every environment variable controlling external connectivity is documented. Your IT security team can inspect the configuration, run network monitoring tools, and confirm that zero outbound connections occur. The system is designed to be independently verifiable, not to require trust in the vendor's claims.
Summary: On-premises AI inference servers operate with zero internet dependency. Models are pre-loaded, inference runs on local NVIDIA H100/H200 GPUs, and all external connectivity features are disabled at the environment variable level. Air-gapped deployment supports SCIF environments, classified networks, NERC CIP-protected systems, and remote field operations. Island Mountain's Summit Series servers start at $75,000 and ship pre-configured for air-gapped operation.
Learn more: Technology & Air-Gap Configuration | Defense Contractors | Energy & Utilities | Government
Talk to Island Mountain about on-premises AI that works without internet. Pre-configured, burn-tested, shipped to your facility.
Request a QuoteOr call directly: 1-801-609-1130