Skip to main content
Version: 2026 R2

Azure AI Foundry

Azure AI Foundry is Microsoft’s platform for accessing a wide range of AI models — from proven models such as GPT-4o, through cost-efficient options, to newer models such as DeepSeek. It offers enterprise-grade reliability, Microsoft technical support, and seamless integration with the Azure ecosystem.

Ready-to-use deployment example

Want to see AI Proxy in action quickly? Use the ready-made demo repository that deploys AI Proxy to Azure Container Instances: WEBCON-AiProxy-GettingStarted.

When to choose Azure AI Foundry​

You are already using Azure:

  • your infrastructure runs in Azure,
  • you want centralized cost and subscription management,
  • you need integration with services such as Azure AD, Key Vault, or Application Insights.

You have enterprise requirements:

  • you require high service availability,
  • you need Microsoft technical support,
  • you must comply with internal corporate policies.

You need flexibility in model selection:

  • you want access to OpenAI models such as GPT-4o and GPT-4o-mini,
  • you also want to use open-source models such as DeepSeek, Llama, or Mistral,
  • you want the option to use Model Router for automatic model selection.

You want better cost control:

  • you need shared TPM limits across multiple deployments,
  • you want detailed usage reporting in Azure Cost Management,
  • you need the ability to define budgets and alerts.

Security and compliance are important to you:

  • data remains in the selected Azure region,
  • the platform supports compliance with GDPR and other regulations,
  • you can use private endpoints and VNET integration.

Requirements​

  • an Active Azure subscription,
  • an Azure AI Foundry workspace with a deployed model,
  • an API Key and endpoint URL.

Step 1: Prepare your Azure environment​

Get access credentials​

  1. Log in to Azure Portal.
  2. Go to your Azure AI Foundry workspace.
  3. In the menu, open Keys and Endpoint.
  4. Copy the following values:
    • Key (API Key),
    • Endpoint OpenAI URL.

Deploy a model​

  1. In your workspace, go to Deployments.
  2. Click Create new deployment.
  3. Select a model, for example GPT-4o, GPT-4o-mini, or another model that supports the Responses API.
  4. Enter a deployment name (e.g., gpt-4o-mini).
  5. Click Create.
Supported models

Azure AI Foundry supports a variety of models, including:

  • GPT models such as gpt-4o, gpt-4o-mini, gpt-35-turbo,
  • Model-router, which automatically selects the most suitable model,
  • DeepSeek and other models that support responses.

Step 2: Configure AI Proxy​

Example aiconfiguration.json​

Choose the example matching your AI Proxy installation version:

{
  "ProviderConnections": {
    "AzureFoundry": {
      "Description": "Azure AI Foundry Connection",
      "Type": "AzureAi",
      "ProviderConfiguration": {
        "ApiKey": "your-azure-api-key-here",
        "Endpoint": "https://your-workspace.openai.azure.com/",
        "UseDefaultAzureCredentials": false
      }
    }
  },
  "ProviderModels": [
    {
      "Id": "11111111-1111-1111-1111-111111111111",
      "ConnectionName": "AzureFoundry",
      "Priority": 100,
      "Name": "Azure GPT-4o-mini",
      "Description": "",
      "Factor": 1.0,
      "TextModel": {
        "ModelName": "gpt-4o-mini"
      },
      "ImageModel": {
        "ModelName": "gpt-4o-mini"
      },
      "AudioModel": {
        "ModelName": "whisper"
      },
      "EmbeddingModel": {
        "ModelName": "text-embedding-3-small"
      }
    }
  ],
  "AiTaskTypesConfiguration": {
    "Concierge": [ "11111111-1111-1111-1111-111111111111" ],
    "EmbeddingGeneration": [ "11111111-1111-1111-1111-111111111111" ]
  }
}
Important

In the ModelName field, enter the deployment name you created in Azure, not the underlying model name. For example, if your deployment is named my-gpt4, use "ModelName": "my-gpt4".

Example docker-compose.yml​

name: aiproxy_containers
services:
  ai-proxy:
    image: webconbps/aiproxy:1.0.0.235
    container_name: ai-proxy
    restart: unless-stopped
    ports:
      - "5298:8080"
      - "7033:8081"
    environment:
      - ASPNETCORE_ENVIRONMENT=Production
      - AppConfiguration__SelfHosted__Certificate__Path=/app/https/certificate.pem
      - Logging__LogLevel__Default=Information
      - Logging__LogLevel__Microsoft=Warning
    volumes:
      - ./certificates/certificate.pem:/app/https/certificate.pem:ro
      - ./aiconfiguration.json:/app/aiconfiguration.json:ro

Step 3: Startup​

# Make sure you have prepared files
# - ./certificates/certificate.pem
# - ./aiconfiguration.json (with filled data)

# Run container
docker-compose up -d

# Check logs
docker-compose logs -f ai-proxy

Troubleshooting​

Error: 401 Unauthorized​

Possible causes:

  • the API key is invalid,
  • the endpoint URL is incorrect.

Solution:

# Check if key and endpoint are correct in aiconfiguration.json
# Verify in Azure Portal > Keys and Endpoint
# Restart container
docker-compose restart ai-proxy

Error: 404 Not Found / Model not found​

Possible causes:

  • the value in ModelName does not match the name of an existing deployment,
  • the specified deployment does not exist or is not active.

Solution:

# Check deployment name in Azure Portal > Deployments
# Make sure deployment has "Succeeded" status
# Update ModelName in aiconfiguration.json
# Restart container
docker-compose restart ai-proxy

Error: 429 Too Many Requests​

Possible cause:

  • the TPM (Tokens Per Minute) limit for the selected deployment has been exceeded.

Solution:

  • wait a moment before retrying,
  • consider increasing the TPM limit for the deployment in Azure.

When creating a deployment, you can choose from models such as:

  • gpt-4o - the latest model from the GPT-4 Optimized family,
  • gpt-4o-mini - a faster and more cost-efficient variant of gpt-4o,
  • gpt-4-turbo - a GPT-4 model with a larger context window,
  • gpt-35-turbo - a GPT-3.5 family model, listed in Azure as gpt-35-turbo,
  • text-embedding-3-small - a model designed for generating embeddings,
  • model-router - a mechanism for automatically selecting the most suitable model,
  • deepseek and other models that support responses.
info

To work correctly with AI Proxy, a model must support the Responses API.