Integrations / Provider Guides

Azure OpenAI

Connect Aurora to Azure-hosted OpenAI deployments with API version and base URL configuration.

Browse docs

Use this provider to route to Azure-hosted OpenAI deployments. Azure requires a base URL and an API version; deployments are configured separately and exposed through Aurora's model list.

Activate

text

AZURE_API_KEY=...
AZURE_BASE_URL=https://<resource>.openai.azure.com
AZURE_API_VERSION=2024-10-21

Restart Aurora. The azure provider appears at /v1/models with your configured deployments.

Default models

Azure does not auto-discover — you list your deployment names in AZURE_MODELS:

text

AZURE_MODELS=my-gpt4o-deployment,my-embedding-deployment

Each ID is the deployment name you created in the Azure portal, not the underlying model name. Your app sends the deployment name as the model string.

Multiple instances

text

AZURE_EAST_API_KEY=...
AZURE_EAST_BASE_URL=https://east.openai.azure.com
AZURE_WEST_API_KEY=...
AZURE_WEST_BASE_URL=https://west.openai.azure.com

Aurora exposes them as azure-east and azure-west.

Custom base URL

The base URL is required. Override per instance as shown above.

Gotchas

Deployment names — model strings sent to Aurora must match the Azure deployment name, not the model name (gpt-4o, etc.). Use aliases if you want to expose stable model names to your app.
API version — pin AZURE_API_VERSION to a known-good version. Microsoft deprecates older versions; check the Azure OpenAI docs for the current list.
Reasoning models — Azure-hosted o-series models require max_completion_tokens rather than max_tokens. Aurora translates automatically.
Function calling / tools — full support, including parallel tool calls.
Prompt caching — supported on Azure-hosted o-series and GPT-4o models. Aurora forwards cache_control and reports cache hits.
Content filtering — Azure applies its own content filters. Aurora's Guardrails run on top.
Quotas — Azure enforces TPM/RPM quotas per deployment. Use Managed API Keys to share quota across your team fairly.

Sample request

text

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AURORA_MASTER_KEY" \
  -d '{
    "model": "azure/my-gpt4o-deployment",
    "messages": [{"role": "user", "content": "Hello from Aurora"}]
  }'

Providers Overview
Aliases — give your app a stable name for your Azure deployment
Pricing — Azure list prices differ from OpenAI list prices; set overrides

Gotchas

Deployment names — model strings sent to Aurora must match the Azure deployment name, not the model name (gpt-4o, etc.). Use aliases if you want to expose stable model names to your app.

API version — pin AZURE_API_VERSION to a known-good version. Microsoft deprecates older versions; check the Azure OpenAI docs for the current list.

Reasoning models — Azure-hosted o-series models require max_completion_tokens rather than max_tokens. Aurora translates automatically.

Function calling / tools — full support, including parallel tool calls.

Prompt caching — supported on Azure-hosted o-series and GPT-4o models. Aurora forwards cache_control and reports cache hits.

Content filtering — Azure applies its own content filters. Aurora's Guardrails run on top.

Quotas — Azure enforces TPM/RPM quotas per deployment. Use Managed API Keys to share quota across your team fairly.

curl http://localhost:8080/v1/chat/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $AURORA_MASTER_KEY" \ -d '{ "model": "azure/my-gpt4o-deployment", "messages": [{"role": "user", "content": "Hello from Aurora"}] }'

Azure OpenAI

Activate

Default models

Multiple instances

Custom base URL

Gotchas

Sample request

Related

Azure OpenAI

Activate

Default models

Multiple instances

Custom base URL

Gotchas

Sample request

Related