Azure OpenAI Configuration

Overview

This guide explains how to enable Azure OpenAI services in your Azure account and deploy AI models for use with AI/Run CodeMie. Azure OpenAI provides access to OpenAI models including GPT-4.1, GPT-5, o1, and o3-series models.

When to Use This Guide

This configuration is required if you plan to use Azure OpenAI models such as GPT-4.1, GPT-5, or any OpenAI models hosted on Azure.

Prerequisites

Before starting, ensure you have:

• Azure Subscription: Active Azure subscription with appropriate permissions
• Resource Group: Existing or ability to create a resource group
• Regional Availability: Check Azure OpenAI model availability by region
• Quota Understanding: Review Azure OpenAI quota and limits

Step 1: Create Azure OpenAI Resource

1.1 Navigate to Azure Portal

1. Go to Azure Portal
2. Sign in with your Azure credentials

1.2 Create Azure OpenAI Service

1. In the search bar, type "Azure OpenAI" and select Azure OpenAI
2. Click Create or Create Azure OpenAI resource
3. Fill in the required configuration:

Basics Tab:

Field	Value	Description
Subscription	Your subscription	Azure subscription to use
Resource group	Select or create new	Logical grouping for resources
Region	Choose region	Select region closest to your users
Name	Unique name	Resource name (globally unique)
Pricing tier	S0 (Standard)	Recommended for production workloads

Network Tab:

Setting	Recommendation	Description
Network	All networks	Allow access from all networks (configure security later)

4. Click Review + create
5. Review settings and click Create
6. Wait for deployment to complete (typically 1-2 minutes)

Step 2: Configure API Access

2.1 Access Resource Keys and Endpoint

1. Once deployment completes, click Go to resource
2. In the left navigation, select Keys and Endpoint
3. Note down the following information:

Information	Where to Find	Purpose
Key 1 or Key 2	Keys and Endpoint section	API authentication
Endpoint	Keys and Endpoint section	API base URL
Location	Overview page	Resource region

2.2 Save Credentials Securely

Store the following information for later use:

# Example credentials (replace with your values)
AZURE_OPENAI_ENDPOINT=https://your-resource-name.openai.azure.com/
AZURE_OPENAI_KEY=your-api-key-here
AZURE_OPENAI_REGION=eastus2

Step 3: Deploy AI Models

3.1 Access Azure AI Foundry Portal

1. Return to your Azure OpenAI resource in Azure Portal
2. On the Overview tab, click Go to Azure AI Foundry portal
3. The Azure AI Foundry portal will open in a new tab

3.2 Navigate to Deployments

1. In Azure AI Foundry portal, locate the left navigation menu
2. Under Shared resources, click Deployments
3. You'll see a list of existing deployments (if any)

3.3 Deploy a Model

1. Click Deploy model or + Create new deployment
2. Select Deploy base model
3. Browse or search for your desired model (e.g., "gpt-4o")
4. Select the model and click Confirm

3.4 Configure Deployment Settings

Setting	Description
Deployment name	Identifier for this deployment
Deployment type	Standard or Global Standard
Model version	Specific model version
Tokens per Minute (TPM)	Rate limit for requests

TPM Rate Limit Guidelines:

• Single Instance: Set to maximum available if this is your only deployment in the region
• Multiple Deployments: Divide quota across deployments based on expected usage
• Check Quota: Navigate to Quotas in left menu to see available capacity

5. Review settings and click Deploy
6. Wait for deployment to complete (typically 30-60 seconds)

3.5 Verify Deployment

1. Once deployed, the model appears in your Deployments list
2. Note the Deployment name - you'll use this in CodeMie configuration
3. Verify Status shows as Succeeded

Step 4: Deploy Additional Models (Optional)

Repeat Step 3 to deploy additional models based on your requirements:

Recommended Models:

• GPT-4o: General-purpose, multimodal reasoning and code generation
• GPT-4o-mini: Cost-effective option for simpler tasks
• GPT-4.1: Advanced reasoning with larger context window
• GPT-5: Latest generation model with enhanced capabilities
• o1/o3-mini: Specialized reasoning models for complex problem-solving
• Embedding models: text-embedding-ada-002 for vector embeddings

Each model deployment allows you to set specific TPM limits and deployment configurations.

Step 5: Configure Regional Deployments (Optional)

For high availability and redundancy:

1. Create Azure OpenAI resources in multiple regions
2. Deploy same models in each region
3. Configure load balancing using LiteLLM Proxy

Quota Management

Understanding Quotas

Azure OpenAI quotas are managed at the region + model level:

• Tokens Per Minute (TPM): Maximum tokens processed per minute
• Requests Per Minute (RPM): Maximum API requests per minute
• Deployment Limits: Maximum number of model deployments per resource

Viewing and Managing Quotas

1. In Azure AI Foundry portal, click Quotas in left navigation
2. View current quota allocation and usage
3. Request quota increases if needed:
   • Click Request quota
   • Fill in justification and requirements
   • Submit request (approval typically takes 2-5 business days)

Security Best Practices

Network Security

1. Enable Private Endpoints:

   • Create private endpoint for Azure OpenAI resource
   • Restrict access to your virtual network

2. Configure Network Rules:

   • Navigate to resource Networking settings
   • Add allowed IP ranges or virtual networks
   • Enable Allow Azure services on the trusted services list