Architecture Overview

System Architecture

Horizons OmniChat is engineered with modularity and flexibility at its core, enabling a robust platform that adapts to diverse deployment scenarios while maintaining consistent functionality and security. Our architecture is designed to scale from individual development environments to enterprise-grade deployments without compromising on performance or security features.

Core Components

The system is thoughtfully divided into four interconnected layers, each designed to handle specific aspects of the platform’s functionality while working in harmony with the others:

graph LR
    A[Client Layer] --> B[Application Layer]
    B --> C[Model Layer]
    B --> D[Persistence Layer]
    
    subgraph "Client Layer"
    A1[Web Interface]
    A2[API Clients]
    end
    
    subgraph "Application Layer"
    B1[Open WebUI]
    B2[Authentication]
    B3[Session Management]
    end
    
    subgraph "Model Layer"
    C1[Ollama]
    C2[Bedrock Gateway]
    end
    
    subgraph "Persistence Layer"
    D1[PostgreSQL/RDS]
    D2[File Storage/EFS]
    end

Component Details

1. Open WebUI

Open WebUI serves as the primary interface and application logic hub of Horizons. Built with modern technologies, it provides a seamless experience for both users and administrators. The component leverages Svelte for the frontend, delivering exceptional performance through its compiled approach, while FastAPI powers the backend with its high-performance asynchronous capabilities. This combination enables:

• A responsive and intuitive chat interface that adapts to various use cases
• Comprehensive model management capabilities for administrators
• Robust user authentication and session management
• Real-time communication through WebSocket support
• RESTful APIs for seamless integration with external systems

2. Ollama

Ollama functions as our local model serving engine, providing sophisticated AI capabilities directly within your infrastructure. It’s designed to optimize model performance while maintaining flexibility:

• Efficient model management with support for multiple model formats
• Intelligent inference optimization based on available hardware
• GPU acceleration capabilities for enhanced performance
• Extensive support for custom models and fine-tuning
• Advanced caching mechanisms for improved response times

3. Bedrock Gateway

The Bedrock Gateway represents our bridge to AWS’s powerful AI capabilities, thoughtfully designed to provide seamless access to cloud-based models while maintaining security and performance:

• Intelligent request routing and load balancing
• Sophisticated request transformation for optimal model interaction
• Robust error handling and retry mechanisms
• Enterprise-grade authentication and security controls
• Comprehensive monitoring and logging capabilities

Deployment Architectures

1. Local Mode

graph LR
    User --> WebUI
    WebUI --> Ollama
    WebUI --> PostgreSQL
    Ollama --> LocalModels

Local mode provides a complete, self-contained environment perfect for development, testing, and privacy-focused deployments. This architecture ensures:

• Complete data sovereignty with all components running locally
• Simplified deployment and maintenance procedures
• Perfect for development and testing environments
• Ideal for organizations with strict data privacy requirements

2. Hybrid Mode

graph LR
    User --> WebUI
    WebUI --> Ollama
    WebUI --> BedrockGateway
    WebUI --> PostgreSQL
    Ollama --> LocalModels
    BedrockGateway --> AWSBedrock

• Components:
  • All Local Mode components
  • Bedrock Gateway container
  • AWS Bedrock integration
• Features:
  • Mixed model support
  • Cloud model access
  • Local infrastructure control
  • Cost optimization

3. AWS Mode

graph LR
    User --> ALB
    ALB --> WebUI-ECS-Fargate
    WebUI-ECS-Fargate --> Ollama-ECS-EC2
    WebUI-ECS-Fargate --> BedrockGateway-ECS-Fargate
    WebUI-ECS-Fargate --> RDS
    Ollama-ECS-EC2 --> EFS
    BedrockGateway-ECS-Fargate --> AWSBedrock

• Components:
  • Application Load Balancer
  • ECS Fargate for WebUI
  • ECS EC2 for Ollama with GPU Optimisation
  • RDS for PostgreSQL
  • EFS for model storage
• Features:
  • Auto-scaling (ENTERPRISE)
  • High availability (ENTERPRISE)
  • Managed services
  • Enterprise security (ENTERPRISE)

Data Flow

1. Chat Request Flow

sequenceDiagram
    participant User
    participant WebUI
    participant ModelBackend
    participant Database
    
    User->>WebUI: Send message
    WebUI->>Database: Log request
    WebUI->>ModelBackend: Forward to model
    ModelBackend->>WebUI: Return response
    WebUI->>Database: Store chat history
    WebUI->>User: Display response

2. Model Management Flow

sequenceDiagram
    participant Admin
    participant WebUI
    participant Ollama
    participant Storage
    
    Admin->>WebUI: Request model install
    WebUI->>Ollama: Download request
    Ollama->>Storage: Save model files
    Storage->>Ollama: Confirm storage
    Ollama->>WebUI: Installation complete
    WebUI->>Admin: Show success

Next Steps

• Review detailed Component Architecture
• Explore Security Architecture
• Check Deployment Options
• Learn about Operations

---

Horizons OmniChat by evereven