Dvina adapts to your infrastructure requirements. Run it on-premise, in your private cloud, or in a hybrid configuration. You control where the AI runs and where your data stays.
Enterprise organizations face unique constraints: data residency requirements, regulatory compliance, security policies, and legacy infrastructure. Some need AI that runs entirely within their data centers. Others require hybrid models where sensitive workloads stay on-premise while general tasks run in the cloud.
Dvina supports all of these scenarios. You choose the deployment model that fits your organization's requirements, not what a vendor dictates.
Deployment Models
On-Premise Deployment
Install and run Dvina entirely within your own data centers. Complete physical control over infrastructure, data, and AI processing.
Private Cloud Deployment
Deploy Dvina in a dedicated cloud environment (AWS VPC, Azure Private Cloud, Google Cloud) with no shared infrastructure.
Hybrid Deployment
Combine on-premise deployment for sensitive workloads with cloud deployment for general use, connected through secure integration.
Air-Gapped Deployment
Run Dvina in completely isolated networks with zero internet connectivity, suitable for classified or highly sensitive environments.
On-Premise Deployment
Organizations with the strictest security and compliance requirements can run Dvina entirely on their own infrastructure.
When On-Premise Makes Sense
Regulatory requirements
Financial institutions, healthcare providers, and government agencies often have policies requiring data to remain within specific physical locations.
Data sovereignty mandates
Countries or sectors requiring data to stay within national borders benefit from complete on-premise control.
Zero external dependencies
Organizations with policies against any cloud usage can run Dvina without external infrastructure.
Air-gapped operations
Defense, intelligence, and critical infrastructure systems that must be completely isolated from the internet.
Legacy system integration
Direct access to internal systems, databases, and APIs that cannot be exposed to external networks.
How It Works
Dvina installs on your infrastructure, connecting to your internal systems, databases, and applications. All AI processing happens on your hardware using local LLMs you select and deploy.
Your team controls:
- Hardware specifications
- Network topology
- Security configurations
- Model selection and updates
- Access policies and authentication
- Data retention and deletion
Dvina provides:
- Installation packages and deployment automation
- Local LLM integration (GPT-OSS, Gemma, DeepSeek, Medical Gemma)
- Connection to your internal tools and databases
- Custom integration development
Private Cloud Deployment
For organizations comfortable with cloud infrastructure but requiring dedicated resources, Dvina offers private cloud deployment.
Benefits Over Multi-Tenant Cloud
Physical isolation
Your Dvina instance runs on dedicated servers. No shared infrastructure with other customers.
Custom network configuration
Define your own network topology, firewall rules, VPN connections, and access controls.
Dedicated resources
No resource contention or "noisy neighbor" issues affecting performance.
Custom security controls
Implement organization-specific security tools, monitoring, and compliance frameworks.
Data residency control
Choose specific geographic regions where your cloud infrastructure operates.
Supported Cloud Providers
- Amazon Web Services (VPC deployment)
- Microsoft Azure (Private Cloud)
- Google Cloud Platform (Private GCP)
- Custom cloud providers upon request
Deployment Architecture
Your private cloud instance includes:
- Dedicated compute resources for AI processing
- Isolated database infrastructure
- Secure connection to your internal systems via VPN or direct connect
- Load balancing and redundancy for high availability
- Automated backups and disaster recovery
Hybrid Deployment
Many organizations need flexibility: sensitive data on-premise, general workloads in the cloud.
Hybrid Architecture
On-Premise Components
- Sensitive data processing with local LLMs
- Connections to internal databases and legacy systems
- High-security workloads requiring physical control
- Regulatory-restricted operations
Cloud Components
- General-purpose AI tasks
- Scalable compute for peak demands
- Public-facing integrations (Gmail, Slack, etc.)
- Collaboration features for distributed teams
Secure Integration
Hybrid deployments connect on-premise and cloud components through encrypted tunnels, ensuring data flows securely between environments while maintaining compliance boundaries.
Use Case Example
A healthcare organization runs Medical Gemma on-premise for patient data analysis, ensuring PHI never leaves their data center. Meanwhile, administrative tasks like scheduling and general communication run in their private cloud for scalability.
Air-Gapped Deployment
For environments requiring complete isolation from external networks, Dvina supports air-gapped deployment.
What Is Air-Gapped?
An air-gapped system has no network connection to the internet or external systems. Data can only enter or leave through physical media or highly controlled, unidirectional connections.
Who Needs Air-Gapped Deployment?
- Defense and military organizations
- Intelligence agencies
- Critical infrastructure operators (power grids, water systems)
- High-security research facilities
- Government classified systems
How Dvina Works Air-Gapped
Initial Installation
Dvina software and local LLM models are delivered via secure physical media or through a controlled one-time connection.
Local LLM Processing
All AI inference happens using models deployed within the air-gapped environment. No external API calls.
Internal System Integration
Dvina connects to databases, applications, and tools within the isolated network.
Updates and Maintenance
Software updates and model improvements are delivered through controlled processes, validated and installed by your security team.
Supported Models for Air-Gapped
- GPT-OSS 120B and 20B
- DeepSeek
- Gemma 3
- Medical Gemma (for healthcare)
- Custom fine-tuned models
Local LLM Deployment
Regardless of deployment model, Dvina supports local language models for complete data sovereignty.
Supported Local Models
General Purpose
- GPT-OSS 120B: Powerful open-source model for complex reasoning, analysis, and generation tasks
- GPT-OSS 20B: Efficient model optimized for resource-constrained environments while maintaining strong performance
- DeepSeek: High-performance model designed for enterprise workloads with efficient inference
- Gemma 3: Google's open-source model family offering flexibility across different scales
Industry-Specific
- Medical Gemma: Healthcare-optimized model trained on medical literature, clinical terminology, and healthcare workflows
- Custom fine-tuned models: Domain-specific models for legal, financial, or other specialized industries
Why Local LLMs Matter
Complete data sovereignty
AI processing happens entirely on your infrastructure. Sensitive data never transmits to external services, ensuring complete control and compliance.
Regulatory compliance
Industries with strict data handling requirements (HIPAA, GDPR, KVKK, BDDK) can use AI while ensuring data remains within compliant infrastructure.
Offline operation
Air-gapped or disconnected environments can run AI without internet connectivity, suitable for classified or high-security operations.
Cost predictability
No per-token API costs. Inference runs on your hardware with fixed infrastructure expenses and predictable scaling.
Customization and fine-tuning
Fine-tune models on your proprietary data for domain-specific improvements, terminology adaptation, and workflow optimization.
Performance optimization
Deploy models on hardware optimized for your specific workloads, from high-throughput GPU clusters to distributed inference systems.
Model Features
GPT-OSS 120B
- Advanced reasoning and analysis capabilities
- Suitable for complex enterprise workflows
- Multi-turn conversation with deep context understanding
- Code generation and technical documentation
- Multilingual support with strong performance
GPT-OSS 20B
- Balanced performance and resource efficiency
- Fast inference for real-time applications
- Strong general-purpose capabilities
- Lower hardware requirements
- Suitable for distributed deployments
DeepSeek
- Optimized inference efficiency
- Strong performance on analytical tasks
- Excellent for structured data processing
- Efficient memory utilization
- Good multilingual capabilities
Gemma 3
- Flexible model family (various sizes)
- Strong instruction following
- Efficient fine-tuning capabilities
- Good balance of performance and speed
- Open-source with active community
Medical Gemma
- Trained on medical literature and clinical data
- Understands medical terminology and abbreviations
- HIPAA-compliant deployment configurations
- Clinical decision support capabilities
- Integration with EHR and medical databases
- Patient data analysis with privacy preservation
Model Selection Guidance
Choose GPT-OSS 120B when:
- Complex reasoning and analysis are required
- Multi-step workflows need deep understanding
- High accuracy is critical
- Infrastructure supports larger models
Choose GPT-OSS 20B when:
- Fast response times are priority
- Distributed deployment is needed
- Hardware resources are limited
- General-purpose tasks without extreme complexity
Choose DeepSeek when:
- Analytical workloads dominate
- Structured data processing is primary use case
- Efficient resource utilization is important
- Cost optimization is a priority
Choose Medical Gemma when:
- Healthcare and clinical applications
- Medical terminology understanding is critical
- HIPAA compliance is required
- Integration with medical systems is needed
Model Deployment Capabilities
Version control and management
Track deployed model versions, test new releases in staging environments, and maintain rollback capabilities for stability.
A/B testing
Deploy multiple model versions simultaneously, route traffic for comparison, and validate performance before full rollout.
Load balancing
Distribute inference requests across multiple model instances for high availability and performance.
Monitoring and observability
Track model performance metrics, inference latency, accuracy indicators, and resource utilization in real-time.
Fine-tuning workflows
Customize models on your proprietary data, adapt to domain-specific terminology, and optimize for your unique use cases.
Implementation Process
Phase 1: Assessment
Work with Dvina's team to understand your infrastructure, security requirements, compliance needs, and use cases. Select appropriate deployment model and local LLMs based on your requirements.
Phase 2: Infrastructure Setup
Deploy Dvina software on your chosen infrastructure, install and configure selected local LLM models, and establish secure connections to internal systems.
Phase 3: Configuration & Integration
Set up authentication and access controls, configure data policies and governance rules, integrate with enterprise tools and databases, and fine-tune models for specific use cases.
Phase 4: Testing & Validation
Conduct security testing, performance benchmarking, compliance validation, and user acceptance testing with pilot groups.
Phase 5: Launch
Train users, execute phased rollout, monitor performance, and optimize configurations based on usage patterns.
Security & Compliance
Infrastructure Security
All deployment models include:
- Encryption at rest (AES-256)
- Encryption in transit (TLS 1.3)
- Network isolation and segmentation
- Intrusion detection and prevention
- Regular security audits
Compliance Support
Dvina supports compliance with:
- GDPR (EU data protection)
- KVKK (Turkish data protection)
- BDDK (Turkish banking regulation)
- HIPAA (US healthcare)
- ISO 27001 (information security)
- SOC 2 Type II (service organization controls)
Audit Capabilities
- Comprehensive logging of all system activities
- User access tracking and reporting
- Data lineage and processing trails
- Compliance report generation
The Bottom Line
Enterprise AI shouldn't force you into someone else's infrastructure. Whether you need on-premise deployment for regulatory compliance, private cloud for dedicated resources, hybrid for flexibility, or air-gapped for maximum security, Dvina adapts to your requirements.
Run local LLMs on your hardware. Connect to your internal systems. Maintain complete control over your data.
Your infrastructure. Your rules. Your AI.