# Air-Gapped AI > Source: https://ibl.ai/resources/glossary/air-gapped-ai > Last updated: 2026-08-19 **Definition:** Air-gapped AI is an AI system deployed on a network with no connection to the public internet, so every model call, prompt and document is processed inside the isolated environment and nothing can be transmitted out. **On ibl.ai you own all the code and the data, run it model-agnostic across any LLM, and pay with no per-seat pricing — so you can deploy anywhere, from your own cloud to a fully air-gapped network.** ## What is Air-Gapped AI? An air gap is a physical and network control, not a policy. There is no outbound route, so data exfiltration is prevented by topology rather than by a vendor's promise or a contractual term. Running AI this way requires open-weight models hosted locally, since any API call to a hosted provider would breach the gap by definition. It also requires that the surrounding platform — orchestration, retrieval, logging, the admin interface — carries no hidden telemetry or license check that needs to phone home. Updates arrive by controlled physical transfer rather than over the network, which makes model and platform update cadence an operational design question rather than an afterthought. ## Why It Matters Air-gapped deployment is the standard for classified government and defense networks, and increasingly for privileged legal matters, clinical research, critical infrastructure control systems, and any environment where a regulator or insurer treats external transmission as unacceptable regardless of encryption. ## Key Characteristics ### No Outbound Route Exists Isolation is enforced by network topology, not configuration. There is no egress path to disable, misconfigure, or re-enable under pressure during an incident. ### Local Model Weights Are Mandatory Every model must be downloaded and hosted inside the boundary. Any hosted-API model is disqualified by definition, which makes open-weight models the practical requirement. ### No Telemetry or Phone-Home Licensing Platform components that check a license server, ship usage analytics, or fetch remote configuration will fail closed or silently break. The stack must be verifiably self-contained. ### Updates by Controlled Physical Transfer Model and platform updates cross the gap on reviewed media, so update cadence, provenance verification and rollback are explicit operational procedures. ### Self-Contained Observability Logs, metrics and audit trails are written and read inside the boundary, since no external monitoring service can be reached to receive them. ### Full Source Access Is Load-Bearing Without the source, an operator cannot verify that a component makes no outbound call, and cannot patch one that does. Air-gapped operation and code ownership are linked. ## Examples - **Defense & Intelligence:** An intelligence agency runs retrieval and summarization over classified holdings on a network with no external route, using locally hosted open-weight models. — *Analysts get AI assistance on classified material with exfiltration prevented by topology rather than by policy or contract.* - **Law Firm:** A law firm processes privileged discovery material on an isolated segment so no privileged content can reach a third-party custodian. — *The firm gets AI-assisted review while its confidentiality duty is satisfied by architecture rather than by a vendor's data-handling terms.* - **Critical Infrastructure:** A utility runs AI over control-system documentation on the operational-technology network, which is isolated from corporate IT and the internet by regulation. — *Operators query decades of maintenance and procedure documents without the OT network gaining an external dependency.* ## Can ibl.ai run fully air-gapped? Yes. ibl.ai is the agentic AI platform where you own all the code and the data, and the entire stack — orchestration, retrieval, agent runtime, admin interface and audit logging — runs inside your perimeter with no outbound connectivity required and no phone-home license check. It is model-agnostic, so you host open-weight models locally on your own GPUs, and carries no per-seat pricing. Because you receive the full source, your security team can verify for itself that no component makes an external call. You can deploy anywhere, from your own cloud to a fully disconnected network. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ## FAQ **Q: What is the difference between air-gapped and on-premise AI?** On-premise means the hardware is yours; it does not by itself mean the system has no internet access. Many on-premise deployments still call hosted model APIs or ship telemetry. Air-gapped means there is no external route at all, which is a stricter and verifiable condition. **Q: Which models can run air-gapped?** Only models whose weights you can download and host — the open-weight families such as Llama, Mistral, Qwen, Gemma and Nemotron. Any model available solely through a hosted API cannot be used, because reaching it would breach the air gap. **Q: How do air-gapped systems receive model and security updates?** Updates cross the boundary on reviewed physical media under a controlled procedure, with provenance verification and a rollback path. This makes update cadence slower and deliberate, which most air-gapped operators treat as a feature rather than a cost. **Q: Does air-gapped AI mean giving up capable models?** Much less than it used to. Open-weight models now handle classification, extraction, retrieval-augmented answering and routine drafting at a quality that covers most enterprise workloads, and a 32GB workstation GPU can serve a 27–34B model quantized. **Q: Can you audit an air-gapped AI deployment?** Yes, and more completely than a hosted one. Every log is written inside your boundary in a schema you control, so the audit trail is an asset you hold rather than a reporting view exposed by a subscription. ## How does ibl.ai approach Air-Gapped AI? **ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.** - **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours. - **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform. - **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ. - **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.