Azure AI Foundry

Azure AI Foundry is the primary provider for OpenDoor, hosting the majority of deployed models.

Overview

OpenDoor connects to Azure AI Foundry via:

• Azure OpenAI endpoint for OpenAI models (GPT-4o, o3-mini, etc.)
• Azure AI Inference API for third-party models (DeepSeek, Kimi, Mistral, etc.)

Configuration

bashCopy
# Required
AZURE_AI_FOUNDRY_ENDPOINT=https://your-account.openai.azure.com/
AZURE_AI_FOUNDRY_KEY=your-key

# Optional - for AI Inference API
AZURE_INFERENCE_ENDPOINT=https://your-ai-services.cognitiveservices.azure.com/
AZURE_INFERENCE_KEY=your-inference-key

# Azure Resource Info
AZURE_SUBSCRIPTION_ID=your-sub-id
AZURE_RESOURCE_GROUP=your-rg
AZURE_OPENAI_ACCOUNT=your-account-name
AZURE_REGION=uksouth

Dynamic Discovery

On startup, the gateway fetches live deployments from the Azure Management API and merges them with the hardcoded catalog.

Supported Models

All models deployed in your Azure OpenAI account are automatically available. The gateway supports both:

• OpenAI-format models (GPT series)
• Publisher-format models (DeepSeek, MoonshotAI, Mistral AI, Cohere, Microsoft)

Fallback Chain

Azure AI Foundry is the first choice for most models. If it fails, the gateway falls back to:

1. OpenAI Direct
2. Anthropic
3. Other configured providers

Deploying New Models

bashCopy
az cognitiveservices account deployment create \
  --name your-account \
  --resource-group your-rg \
  --deployment-name ModelName \
  --model-name ModelName \
  --model-version "1" \
  --model-format PublisherFormat \
  --sku-capacity 1 --sku-name GlobalStandard