Azure OpenAI setup (conceptual)

Azure OpenAI’s inference is accessed via data plane endpoints. Calls are versioned using an api-version query parameter in YYYY-MM-DD format. The Microsoft Learn reference describes the GA inference spec as 2024-10-21.

Reference: Microsoft Learn — Azure OpenAI in Microsoft Foundry Models REST API reference.

Endpoint shape (Chat Completions)

For Azure OpenAI resources, the chat completions route follows the pattern:

POST https://{endpoint}/openai/deployments/{deployment-id}/chat/completions?api-version=2024-10-21

and is authenticated via api-key or Microsoft Entra ID.

What this playbook assumes

• You already have an Azure OpenAI resource + a deployed model.
• You will place your deployment name in one place: AZURE_OPENAI_DEPLOYMENT in .env.

Next: Authentication & Endpoints.