Air-Gapped

General definition

Air-Gapped describes a computer or network that has no network path to the internet or to any less trusted network. The gap is literal: no cable, no Wi-Fi, no VPN. Data crosses the boundary only through controlled physical transfer, such as removable media, a data diode that allows one-way flow, or a supervised import station. It is the strongest network isolation available and is used where the cost of a breach is extreme: defence and intelligence, industrial control systems and critical infrastructure, some hospital clinical networks, trading and settlement systems, and research environments handling sensitive datasets.

Running chat and AI software inside an air gap changes almost every assumption a cloud product makes:

• Installation must be offline: container images, packages and licences delivered as files and loaded from an internal registry or mirror, with checksums verified on import
• Updates follow the same path on a schedule, so releases need to be self-contained and reversible
• Certificates and identity come from an internal certificate authority and directory, not a public CA or a hosted identity provider
• AI models must run locally. There is no route to a cloud LLM API, so inference uses a local LLM served by tools such as Ollama or vLLM on hardware inside the boundary, with model weights imported like any other artefact
• Push notifications through Firebase Cloud Messaging or Apple’s APNs are not possible, because both are public cloud services. Notification has to come from a persistent connection inside the network or from a device channel the organisation manages
• Telemetry and licence checks that call home must be absent or switchable off

Air-gapped is the far end of a spectrum. Many organisations do not need a physical gap but do need no internet egress: the system lives on a private network, reaches nothing outside it, and is reached only through controlled entry points. That requirement is met by an on-premise or isolated private-cloud deployment, and the same software design rules apply, which is why vendors that support one usually support the other.

In the Ethora ecosystem

Ethora’s self-hosted deployment is a set of Docker containers (the XMPP server, the Node.js API, MongoDB, file storage and monitoring) that run on infrastructure you control, and its AI agents can point at any OpenAI-compatible endpoint, including a private LLM served by Ollama or vLLM on the same network. Architecturally that means a deployment can operate with no internet egress: users are provisioned from your systems, messages and files stay on your servers, and model inference happens on your hardware. The self-hosted LLM AI agent page describes the model side.

What an air-gapped install still requires from your side is the transport of images, packages and model weights across the boundary, an internal certificate authority, and a decision about notifications, since FCM and APNs are unavailable and clients on the isolated network rely on their live connection instead. Ethora does not claim a certified air-gap product; it claims an architecture with no built-in dependency on Ethora-operated services once deployed, which is the property an isolation review looks for. Talk to the team about a specific environment.