xr-ai-accelerator

XrAiModelManager

The XrAiModelManager class is a MonoBehaviour component that manages AI model configurations, API keys, and workflow-specific properties. It provides a centralized system for storing and retrieving model settings across different AI providers and workflows.

Class Declaration

public class XrAiModelManager : MonoBehaviour

Data Structures

XrAiProperty

Represents a key-value pair for configuration properties.

[System.Serializable]
public class XrAiProperty
{
    public string key;
    public string value;
}

XrAiSection

Groups related properties under a section name.

[System.Serializable]
public class XrAiSection
{
    public string sectionName;
    public List<XrAiProperty> properties = new List<XrAiProperty>();
}

XrAiModelData

Container for all configuration sections.

[System.Serializable]
public class XrAiModelData
{
    public List<XrAiSection> sections = new List<XrAiSection>();
}

Constants

File Paths

private const string CONFIG_FILE_PATH = "XrAiModelConfig";
private const string API_KEYS_FILE_PATH = "XrAiApiKeys";

Workflow Types

public const string WORKFLOW_IMAGE_TO_IMAGE = "ImageToImage";
public const string WORKFLOW_IMAGE_TO_TEXT = "ImageToText";
public const string WORKFLOW_TEXT_TO_IMAGE = "TextToImage";
public const string WORKFLOW_SPEECH_TO_TEXT = "SpeechToText";
public const string WORKFLOW_IMAGE_TO_3D = "ImageTo3d";
public const string WORKFLOW_OBJECT_DETECTOR = "ObjectDetector";
public const string WORKFLOW_TEXT_TO_SPEECH = "TextToSpeech";

Properties

ModelData

Gets the main configuration data, loading from file if necessary.

public XrAiModelData ModelData { get; }

ApiKeysData

Gets the API keys data, separated from main configuration for security.

public XrAiModelData ApiKeysData { get; }

Methods

LoadFromFile

Loads configuration data from Resources folder files.

public void LoadFromFile()

SaveToFile

Saves configuration data to Resources folder files.

public void SaveToFile()

GetGlobalProperties

Retrieves all global properties for a section, including API keys.

public Dictionary<string, string> GetGlobalProperties(string sectionName)

GetWorkflowProperties

Retrieves workflow-specific properties for a section.

public Dictionary<string, string> GetWorkflowProperties(string sectionName, string workflowName)

GetGlobalProperty

Retrieves a single global property value.

public string GetGlobalProperty(string sectionName, string key, string defaultValue = "")

GetWorkflowProperty

Retrieves a single workflow-specific property value.

public string GetWorkflowProperty(string sectionName, string workflowName, string key, string defaultValue = "")

Usage Examples

Basic Setup

public class AIConfigurationManager : MonoBehaviour
{
    private XrAiModelManager modelManager;
    
    void Start()
    {
        modelManager = GetComponent<XrAiModelManager>();
        ConfigureAIModels();
    }
    
    private void ConfigureAIModels()
    {
        // Get OpenAI configuration
        var openAIConfig = modelManager.GetGlobalProperties("OpenAI");
        string apiKey = openAIConfig.GetValueOrDefault("apiKey", "");
        
        if (string.IsNullOrEmpty(apiKey))
        {
            Debug.LogWarning("OpenAI API key not configured");
        }
        
        // Get workflow-specific settings
        string imageToTextModel = modelManager.GetWorkflowProperty(
            "OpenAI", 
            XrAiModelManager.WORKFLOW_IMAGE_TO_TEXT, 
            "model", 
            "gpt-4-vision-preview"
        );
    }
}

Dynamic Configuration

public class RuntimeConfiguration : MonoBehaviour
{
    private XrAiModelManager modelManager;
    
    public void SetupProviderConfiguration(string provider, Dictionary<string, string> config)
    {
        modelManager = GetComponent<XrAiModelManager>();
        
        // Update configuration at runtime
        foreach (var setting in config)
        {
            SetGlobalProperty(provider, setting.Key, setting.Value);
        }
        
        // Save changes
        modelManager.SaveToFile();
    }
    
    private void SetGlobalProperty(string section, string key, string value)
    {
        // This would require extending the manager to support runtime updates
        // or directly modifying the ModelData structure
    }
}

Workflow-Specific Configuration

public class WorkflowConfigurationExample : MonoBehaviour
{
    private XrAiModelManager modelManager;
    
    public async Task<string> ProcessImageWithGroq(Texture2D image)
    {
        modelManager = GetComponent<XrAiModelManager>();
        
        // Get Groq configuration for Image-to-Text workflow
        var groqConfig = modelManager.GetGlobalProperties("Groq");
        var workflowConfig = modelManager.GetWorkflowProperties("Groq", XrAiModelManager.WORKFLOW_IMAGE_TO_TEXT);
        
        // Load the model with configuration
        var imageToText = XrAiFactory.LoadImageToText("Groq", groqConfig);
        
        // Execute with workflow-specific options
        byte[] imageData = XrAiImageHelper.EncodeTexture(image, "image/jpeg");
        var result = await imageToText.Execute(imageData, "image/jpeg", workflowConfig);
        
        return result.IsSuccess ? result.Data : null;
    }
}

Configuration File Structure

Main Configuration (XrAiModelConfig.txt)

{
  "sections": [
    {
      "sectionName": "Groq",
      "properties": []
    },
    {
      "sectionName": "Groq.ImageToText",
      "properties": [
        {
          "key": "model",
          "value": "llama-vision-free"
        },
        {
          "key": "prompt",
          "value": "Describe what you see in this image."
        }
      ]
    },
    {
      "sectionName": "OpenAI.SpeechToText",
      "properties": [
        {
          "key": "model",
          "value": "whisper-1"
        }
      ]
    }
  ]
}

API Keys Configuration (XrAiApiKeys.txt)

{
  "sections": [
    {
      "sectionName": "Groq",
      "properties": [
        {
          "key": "apiKey",
          "value": "your-groq-api-key"
        }
      ]
    },
    {
      "sectionName": "OpenAI",
      "properties": [
        {
          "key": "apiKey",
          "value": "your-openai-api-key"
        }
      ]
    }
  ]
}

Predefined Sections and Properties

The manager automatically initializes configuration for known providers:

Supported Providers

Nvidia: Image-to-Text
OpenAI: Speech-to-Text, Text-to-Image, Image-to-Image, Text-to-Speech
StabilityAi: Image-to-3D
Google: Image-to-Text, Object Detection
Groq: Image-to-Text
Roboflow: Object Detection
RoboflowLocal: Object Detection

Auto-Initialized Properties

Each provider section automatically includes:

API key property (stored separately)
Workflow-specific properties based on capabilities

Security Considerations

API Key Separation

API keys are stored in a separate file (XrAiApiKeys.txt)
Main configuration can be shared without exposing sensitive data
Consider encrypting the API keys file for production

File Location

Configuration files are stored in the Resources folder
Files are included in builds but accessible at runtime
Consider using persistent data path for user-specific configurations

Integration with AI Factory

public class FactoryIntegration : MonoBehaviour
{
    private XrAiModelManager modelManager;
    
    public IXrAiImageToText CreateConfiguredImageToText(string provider)
    {
        modelManager = GetComponent<XrAiModelManager>();
        
        // Get provider configuration
        var config = modelManager.GetGlobalProperties(provider);
        
        // Load model with configuration
        return XrAiFactory.LoadImageToText(provider, config);
    }
    
    public async Task<XrAiResult<string>> ExecuteWithWorkflowConfig(
        IXrAiImageToText model, 
        byte[] imageData, 
        string provider)
    {
        // Get workflow-specific options
        var workflowOptions = modelManager.GetWorkflowProperties(
            provider, 
            XrAiModelManager.WORKFLOW_IMAGE_TO_TEXT
        );
        
        return await model.Execute(imageData, "image/jpeg", workflowOptions);
    }
}

Editor Integration

The manager supports Unity Editor integration:

#if UNITY_EDITOR
// Auto-refresh assets when saving configuration
UnityEditor.AssetDatabase.Refresh();
#endif

Best Practices

Separate Concerns: Keep API keys separate from workflow configuration
Default Values: Always provide sensible defaults for configuration properties
Validation: Validate configuration before using with AI models
Environment-Specific: Consider different configurations for development/production
Documentation: Document all configuration properties and their purposes

Implementation Notes

Inherits from MonoBehaviour for Unity integration
Uses JSON serialization for configuration persistence
Automatically initializes predefined provider configurations
Separates API keys from general configuration for security
Supports both global and workflow-specific property management
Integrates with Unity’s Resources system for easy asset management