Air-Gapped Deployment

Zero Network Dependency

Most AI platforms treat offline mode as a fallback — a degraded state that kicks in when the connection drops. Guaardvark takes the opposite approach. It was designed from the ground up with no cloud callbacks whatsoever. There is no telemetry phoning home, no license server validation, no automatic update checks, and no CDN dependencies baked into the runtime. The application does not assume a network exists, and it never attempts outbound connections.

Every dependency that Guaardvark requires is either bundled directly into the installation package or installable from local media such as a USB drive or an internal repository mirror. The AI models, the inference engine, the database, the cache layer, the frontend assets — all of it ships as a single, self-contained unit. Once installed, the system runs indefinitely without network access. There is no expiration timer, no grace period, and no degraded feature set. What you install is what you get, permanently.

This architecture matters because air-gapped environments cannot tolerate even a single unauthorized outbound packet. A platform that tries to resolve a DNS name or reach an API endpoint will trigger intrusion detection systems, violate security policies, and erode trust in the tool. Guaardvark eliminates that risk entirely by never generating outbound traffic in the first place. Your security team can verify this with a packet capture during installation and operation — the network interface stays silent.

Ideal Environments

Air-gapped deployments are mandated across some of the most security-sensitive environments in the world. Guaardvark is built to serve them all.

• SCIFs — Sensitive Compartmented Information Facilities require absolute network isolation. Guaardvark provides full AI capability inside these rooms without compromising the physical or logical security boundary.
• Classified government networks — Systems operating on SIPRNet, JWICS, and other classified enclaves cannot connect to the public internet. Guaardvark runs natively on these networks, giving analysts and operators access to local LLMs, document search, and AI-assisted workflows without any cross-domain risk.
• Air-gapped industrial control systems — Critical infrastructure operators in energy, manufacturing, and utilities maintain strict separation between IT and OT networks. Guaardvark can be deployed inside the OT perimeter to provide AI-powered monitoring, anomaly analysis, and document retrieval without bridging the air gap.
• Remote field deployments — Military forward operating bases, research stations, maritime vessels, and remote mining or drilling operations often lack reliable connectivity. Guaardvark operates fully offline, providing AI assistance in locations where a satellite uplink is the only option and bandwidth is measured in kilobits.
• Healthcare facilities with strict data isolation — Hospitals and research institutions handling protected health information can deploy Guaardvark on isolated networks to ensure patient data never leaves the facility, simplifying HIPAA compliance and data governance.
• Financial trading floors with restricted internet access — High-frequency trading desks and risk analysis teams that operate on locked-down networks can leverage local AI for research, summarization, and code review without introducing external data exfiltration vectors.

How It Works

Deploying Guaardvark in an air-gapped environment follows a straightforward process. Every component is designed for offline installation and operation from the very first step.

Compliance Benefits

Running AI in regulated environments introduces compliance complexity that most cloud-based platforms struggle to address. Guaardvark simplifies your compliance posture by keeping everything within your security boundary.

• Data residency guarantees — Your data never crosses network boundaries. Every prompt, every response, every generated image, and every search result stays on the local machine. There is no ambiguity about where data is processed or stored.
• Full audit trail — Every AI interaction is logged locally with timestamps, user identifiers, model versions, and complete input/output pairs. These logs are stored in the local PostgreSQL database and can be exported for compliance review without involving any third party.
• No third-party data processing agreements — Because no data leaves your infrastructure, you do not need Data Processing Agreements, Business Associate Agreements, or Standard Contractual Clauses with cloud AI providers. This eliminates an entire category of legal and procurement overhead.
• Simplified regulatory compliance — Guaardvark's architecture directly supports FISMA, FedRAMP, HIPAA, GDPR, and SOC 2 requirements by ensuring data sovereignty, access control, and auditability are built into the platform rather than bolted on through configuration.
• AI capability without compromise — The traditional trade-off has been: adopt AI and accept cloud dependency, or maintain your security boundary and forgo AI entirely. Guaardvark eliminates that trade-off. You get a full-featured AI platform that strengthens rather than weakens your security posture.

Multi-Device Deployment

A single air-gapped machine running AI is valuable. A coordinated fleet of machines running AI across an isolated network is transformational. Guaardvark's Interconnector system enables exactly that.

The Interconnector creates a local AI mesh within your air-gapped network. Multiple Guaardvark instances discover each other on the LAN and form a synchronized cluster — no internet gateway required. You can sync models between devices so that a newly provisioned workstation receives the same LLM configuration as every other node. Share prompt templates, system configurations, and RAG document collections across the mesh with a single broadcast command.

Architecture is flexible: deploy a centralized hub that serves as the authoritative model repository and configuration source, or run a fully distributed peer-to-peer topology where each node is equal. The Interconnector handles conflict resolution, version tracking, and incremental synchronization automatically. Updates to models or configurations propagate across the local network without any machine needing to contact the outside world.

For organizations scaling AI across a classified enclave or an isolated facility, this means deploying once to a single device and then replicating the full AI stack to dozens or hundreds of endpoints with minimal manual intervention — all within the air gap.

What You Get

An air-gapped deployment of Guaardvark is not a stripped-down version of the platform. You get the complete feature set, running entirely on local hardware, with no capability sacrificed for the sake of disconnection.

• Full AI chat with local LLMs — Conversational AI powered by models like Llama, Mistral, Qwen, and DeepSeek running through Ollama. Switch between models at runtime based on the task at hand.
• Image and video generation — Create images with Stable Diffusion and generate videos with Wan2.2 or CogVideoX, all processed locally on your GPU. No cloud rendering, no API limits, no per-image fees.
• Document search and RAG — Ingest documents, build vector indexes, and run retrieval-augmented generation queries against your own knowledge base. BM25 and dense vector retrieval work together to surface the most relevant information.
• AI agents for task automation — ReACT-loop agents that reason, plan, select tools, execute actions, and iterate autonomously. Automate research, code generation, document analysis, and multi-step workflows without human intervention.
• Voice interaction — Speak to Guaardvark using Whisper.cpp for speech-to-text and Piper TTS for text-to-speech. Fully offline voice interaction with no audio data leaving the device.
• Code assistance — Integrated Monaco editor with AI-powered code completion, review, and refactoring. Write and iterate on code within the platform, with the AI understanding your full codebase context.

All of this runs without a single packet leaving your network. The air gap is maintained at every layer of the stack, from inference to storage to user interface.