AI Without the Internet

Some environments can’t connect to the internet. Air-gapped networks, classified systems, remote locations, secure facilities—they need AI capabilities without cloud connectivity.

Calliope supports true air-gapped deployment.

Why Air-Gap?

Organizations require air-gapped AI for:

Classified environments: Government and defense systems handling classified data

Regulatory requirements: Industries with strict data isolation rules

Critical infrastructure: Systems where internet exposure creates unacceptable risk

Remote locations: Facilities without reliable connectivity

How Air-Gapped AI Works

In an air-gapped environment:

1. Local deployment: Calliope runs in your isolated network
2. Local models: AI models run via Ollama on your hardware
3. Local data: Everything stays within your perimeter
4. No external calls: Zero internet connectivity required

Local Model Options

Air-gapped deployments use local models:

Llama family:

• Llama 3 (various sizes)
• CodeLlama for development
• Fine-tuned variants

Mistral family:

• Mistral 7B
• Mixtral (mixture of experts)

Other options:

• Custom fine-tuned models
• Approved models for your environment

Models run on your hardware via Ollama.

Hardware Requirements

Local AI requires local compute:

Minimum (7B models):

• 16GB RAM
• Modern CPU
• 50GB storage

Recommended (13B models):

• 32GB RAM
• GPU with 12GB+ VRAM
• 100GB storage

Production (70B+ models):

• Multiple GPUs
• Significant RAM
• Fast storage

Larger models = better quality = more resources.

Deployment Architecture

Air-gapped Calliope deployment:

[Your Air-Gapped Network]
├── Calliope Platform
│   ├── AI Lab
│   ├── AI IDE
│   ├── Chat Studio
│   └── Deep Agent
├── Ollama (Model Server)
│   ├── Llama 3
│   ├── CodeLlama
│   └── Mistral
├── Local Database
└── Document Storage

Everything runs locally. Nothing leaves.

Getting Software Into Air-Gapped Environments

Typical air-gap deployment process:

1. Download packages on connected system
2. Security scan of all packages
3. Transfer media (approved method)
4. Install in air-gapped environment
5. Configure for local operation
6. Test functionality

We provide offline installation packages.

Model Updates

Updating models in air-gapped environments:

1. New model version available
2. Security review of model package
3. Approval for air-gap transfer
4. Transfer via approved method
5. Update local Ollama
6. Validate functionality

Less frequent than cloud, but manageable.

Performance Expectations

Air-gapped AI performance depends on:

Model size: Smaller models are faster but less capable

Hardware: GPU acceleration dramatically improves speed

Workload: Batch vs. interactive affects perceived performance

Set expectations based on your hardware and model choices.

Security Considerations

Air-gapped deployment provides:

• Complete data isolation
• No external data exposure
• Full control over software
• Auditable operations

But also requires:

• Local security management
• Update planning
• Hardware maintenance
• Operational expertise

Use Cases

Development environments: Secure coding assistance without code exposure

Analysis: Data analysis on classified datasets

Research: AI-assisted research on sensitive data

Operations: Operational support for critical systems

The Air-Gap Checklist

For air-gapped AI deployment:

• Hardware sized for target models
• Installation media prepared and scanned
• Network isolation verified
• Local models selected and approved
• Update process defined
• Operations team trained
• Security review completed

AI capabilities, complete isolation.

Discuss air-gapped deployment →