Skip to main content

Azure OpenAI Integration

Azure OpenAI Setup​

Azure OpenAI Service provides REST API access to OpenAI's powerful language models including GPT-4, GPT-3.5-Turbo, and Embeddings model series. To use Azure OpenAI, follow these steps:

  1. Access Azure OpenAI Studio

    • Go to Azure OpenAI Studio
    • Sign in with your Azure account
    • You'll see the "Azure OpenAI resources" page
    • Click the Create new Azure OpenAI resource link in the left part of the header

  2. Create Your Resource

  3. Access Your Resource

    • Once created, your resource will appear in the resources list
    • You'll see details including:
      • Resource name
      • Subscription
      • Region
      • Pricing tier
      • Endpoint
      • Key
    • Click the resource name to start working with your Azure OpenAI deployment

  4. Get Integration Details

    • In Azure AI Studio, click the resource selector in the top-right corner
    • A sidebar will appear with your resource details
    • You'll need:
      • Azure OpenAI Resource Name: Found under "Azure OpenAI or AI services Resource"
      • API Key: Found under "Key" (click the eye icon to reveal)
    • These values are required for setting up the BotDojo integration

Creating a New Azure OpenAI Resource​

When creating a new Azure OpenAI resource, you'll need to complete a multi-step form. Here's how to fill out each section:

Basics Tab​

  1. Project Details
    • Subscription: Select your Azure subscription from the dropdown
    • Resource group: Either select an existing group or click "Create new"
      • If creating new, we recommend naming it clearly (e.g., "openai-resources")
      • We recommend creating the resource group in East US or East US 2 for best model availability
    • Name: Choose a unique name for your resource (this will be part of your API endpoint)
    • Region: Select East US or East US 2 for best model availability
    • Pricing tier: Standard S0 (default)

Network Tab​

  1. Network Access
    • We recommend selecting one of these options:
      • All networks (Simplest): Allows access from any internet connection
      • Selected networks (More secure): Limits access to specific networks
        • Note: This option requires additional network configuration
        • Contact us to schedule a 1-on-1 meeting for setup assistance
    • We don't recommend "Disabled" for initial setup

Tags Tab​

  • Tags help you organize and track Azure resources
  • Common tags to consider:
    • Environment: (e.g., "production", "development", "staging")
    • Project: Your project name
    • Department: Your team or department
    • Owner: Your team's contact
    • CostCenter: For billing purposes
  • While optional, adding tags now makes resource management easier later

Review + submit​

  1. Carefully review all settings:
    • Verify your subscription and resource group
    • Confirm the resource name is correct (can't be changed later)
    • Double-check the region selection
    • Verify network access settings
    • Review any tags you've added
  2. Validation
    • Azure will automatically validate your configuration
    • Address any validation errors if they appear
  3. Click "Create"
    • The deployment process will begin
    • Wait for deployment to complete (usually takes 3-5 minutes)
  4. After Deployment
    • You'll receive a notification when deployment is complete
    • Click "Go to resource" to access your new Azure OpenAI resource
    • You can now proceed with model deployments

Deploy Models in Azure OpenAI​

Before using models, you need to deploy them in your Azure OpenAI resource. You can do this through the Azure OpenAI Studio:

  1. Access Azure OpenAI Studio

    • Go to Azure OpenAI Studio
    • Sign in with your Azure account
    • Select your Azure OpenAI resource from the dropdown in the top navigation

  2. Deploy a Model

    • In the left sidebar, click Deployments
    • Click + Create new deployment
    • Select a model from the dropdown
    • IMPORTANT: The deployment name MUST EXACTLY match BotDojo's expected model names:
      • gpt-4o for GPT-4 Omni
      • gpt-4o-mini for GPT-4 Omni Mini
      • gpt-4 for GPT-4
      • gpt-35-turbo for GPT-3.5 Turbo
      • text-embedding-3-large for embeddings (large)
      • text-embedding-3-small for embeddings (small)
    • Any deviation from these exact names will cause the integration to fail
    • Click Create

  3. Verify Deployment

    • After creation, check that the deployment name exactly matches the expected name
    • The model will be available for use once the deployment status shows as "Succeeded"

⚠️ Critical Note: Model deployment names are case-sensitive and must exactly match the names listed above. Using different names (even slightly different) will prevent BotDojo from finding and using the models.

Model Deployment Best Practices​

  1. Capacity Planning

    • Start with lower TPM (tokens per minute) limits
    • Monitor usage and adjust as needed
    • Consider deploying multiple models for different use cases

  2. Region Selection

    • Check model availability in your region

  3. Security Considerations

    • Store API keys securely
    • Use Azure Key Vault for production deployments
    • Implement proper access controls
    • Monitor usage for unusual patterns

Usage Quotas and Limits​

  • Each deployment has TPM (tokens per minute) limits
  • Quotas can be increased through Azure support requests
  • Monitor usage through Azure metrics
  • Set up alerts for quota approaching limits

Cost Management​

  • Costs are based on token usage
  • Different models have different pricing tiers
  • Monitor costs through Azure Cost Management
  • Set up budgets and alerts to prevent overruns

Troubleshooting​

Common issues and solutions:

  • Rate Limiting: Implement exponential backoff
  • Token Limits: Check model context windows
  • API Errors: Verify credentials and endpoints
  • Deployment Issues: Ensure model is properly deployed

For more detailed information, refer to the Azure OpenAI documentation.