Azure OpenAI

Provider value: azure

Azure OpenAI runs the same models as OpenAI but through your own Azure resource, with Azure's authentication and compliance guarantees. ra uses AzureOpenAI from the openai SDK and supports both API key and DefaultAzureCredential authentication.

Setup

Option 1: API key

export RA_AZURE_ENDPOINT=https://myresource.openai.azure.com/
export RA_AZURE_DEPLOYMENT=my-gpt4o
export RA_AZURE_API_KEY=your-azure-api-key
ra --provider azure "Hello"

Option 2: DefaultAzureCredential (recommended for Azure-hosted workloads)

When RA_AZURE_API_KEY is not set, ra falls back to DefaultAzureCredential, which tries these in order:

1. AZURE_CLIENT_ID / AZURE_CLIENT_SECRET / AZURE_TENANT_ID env vars
2. Workload identity (AKS)
3. Managed identity (Azure VMs, App Service, Container Apps)
4. Azure CLI (az login)
5. Azure Developer CLI (azd auth login)

export RA_AZURE_ENDPOINT=https://myresource.openai.azure.com/
export RA_AZURE_DEPLOYMENT=my-gpt4o
ra --provider azure "Hello"

Environment variables

Credentials are env-only — never passed as CLI flags to keep them out of shell history.

Variable	Required	Description
RA_AZURE_ENDPOINT	Yes	Azure OpenAI resource endpoint
RA_AZURE_DEPLOYMENT	Yes	Deployment name (as configured in Azure AI Studio)
RA_AZURE_API_KEY	No	API key auth. Omit to use DefaultAzureCredential
RA_AZURE_API_VERSION	No	API version, e.g. 2024-12-01-preview

CLI flags

Non-sensitive options can also be set via CLI:

Flag	Description
--azure-endpoint	Azure resource endpoint
--azure-deployment	Deployment name

Config file

provider: azure
providers:
  azure:
    endpoint: https://myresource.openai.azure.com/
    deployment: my-gpt4o
    apiVersion: 2024-12-01-preview

Deployment vs model

In Azure OpenAI, you deploy a model under a name you choose (e.g. my-gpt4o). ra uses the deployment name for all API calls — the --model flag is ignored when using Azure. Set your deployment via RA_AZURE_DEPLOYMENT or --azure-deployment.

Extended thinking

Supported levels: low, medium, high (requires a reasoning-capable deployment).

ra --provider azure --thinking medium "Analyze this architecture"

See also

• Context Control — extended thinking details
• Configuration — provider credentials reference