Most AI systems need an internet connection. They call external APIs, download model updates, and send data to cloud servers for processing. For some organisations, that's a non-starter. When you're handling classified information, sensitive research, or data that absolutely cannot leave your facility, you need AI that works with zero network connectivity.
That's what airgapped AI delivers.
What Is an Air Gap?
An air gap is a physical separation between a computer system and any external network. There's no Wi-Fi, no ethernet cable, no Bluetooth — no connection of any kind to the outside world. Data can only enter or leave through controlled physical media (USB drives, optical discs) that go through security screening.
Air gaps have been used for decades in military, intelligence, and critical infrastructure environments. Airgapped AI brings modern language model capabilities into these isolated environments.
Who Needs Airgapped AI?
- Defence and intelligence — classified information processing
- Government agencies — sensitive policy and citizen data
- Research facilities — proprietary or pre-patent research
- Critical infrastructure — power grids, water systems, transport networks
- Financial institutions — trading algorithms and market-sensitive data
- Legal firms — high-profile cases with extreme confidentiality requirements
- Healthcare — genomic data, clinical trials, sensitive patient cohorts
How Airgapped AI Works
1. Hardware Provisioning
The AI runs on dedicated hardware within your secure facility. This typically means GPU-equipped servers (for model inference) plus storage for your data and the model weights. The hardware is configured and tested before being physically installed in the airgapped environment.
2. Model Installation
The language model (e.g., LLaMA or similar open-source model) is loaded onto the hardware via verified physical media. The model weights, configuration, and all dependencies are packaged into a self-contained deployment that needs nothing from the internet.
3. Data Loading
Your organisation's data — documents, databases, knowledge bases — is loaded through the same controlled physical transfer process. This data is used for retrieval-augmented generation (RAG) or fine-tuning, giving the AI knowledge of your specific domain.
4. MCP Server Configuration
MCP servers are configured to connect the AI to local databases and file systems within the airgapped network. All the same integration capabilities are available — just without any external network access.
5. User Access
Users access the AI through a web interface on the internal network. The experience is identical to using any other AI chat tool — the air gap is invisible to end users.
What You Can Do with Airgapped AI
Everything a connected AI can do, minus internet access:
- Natural language queries against your internal documents
- Document drafting, summarisation, and analysis
- Code generation and review
- Data analysis and report generation
- Translation and language processing
- Knowledge base search and Q&A
The air gap restricts network access, not AI capability. Your team gets the same productivity benefits — just without the security risks.
Updating an Airgapped System
The main operational difference with airgapped AI is updates. Model improvements, security patches, and new data must be transferred via physical media through your existing security protocols. We package updates into verified, signed bundles that your security team can review before installation.
Update frequency depends on your needs — some organisations update monthly, others quarterly. The AI remains fully functional between updates.
Performance Considerations
Airgapped AI performance depends on the hardware you allocate. Modern open-source models run efficiently on enterprise GPU hardware:
- Response time: Typically 1-5 seconds for standard queries
- Concurrent users: Depends on GPU count — a single high-end GPU handles 5-10 simultaneous users comfortably
- Model quality: Open-source models like LLaMA now rival commercial offerings for most business tasks
Getting Started
Deploying airgapped AI requires careful planning around hardware, security protocols, and data transfer procedures. We work with your security team to design a deployment that meets your specific requirements.
Contact us to discuss airgapped AI deployment for your organisation.