Capabilities

@anvia/azure is a thin Azure configuration layer over the OpenAI adapters. Completion, embedding, image, speech, and transcription behavior is the same normalized mapping that @anvia/openai provides, applied to Azure deployments.

Capability	Support
Responses completion and streaming	Yes (api: 'responses')
Chat Completions and streaming	Yes (default)
Local Anvia tools and tool choice	Yes
Structured output	Where the deployment supports it
Reasoning content	Where the deployment supports it
Embeddings	Yes
Image generation, speech, transcription	Where the deployment supports them
Model listing	Endpoint model list, as azure-openai
API-key authentication	Yes
Microsoft Entra token provider	Yes
Injected OpenAI SDK client	Yes
Azure v1 endpoints (/openai/v1/)	Yes
Foundry project endpoints	Yes, through baseUrl

Which APIs, tools, media options, and listing results work depends on your Azure endpoint and deployment. listModels() calls the endpoint's model listing API; it does not enumerate Azure Resource Manager deployments.

Azure Responses behavior

The Responses model extends the shared adapter that @anvia/openai/adapters exports and adds two Azure-specific behaviors:

• Message item types. Conversation messages are sent with an explicit type: "message", which Foundry project endpoints require and resource endpoints accept. Function-call and tool-result items keep their own shapes.
• Terminal tool-argument names. Azure streams may omit the function name from the terminal response.function_call_arguments.done event. The model resolves it from the earlier function-call item with the same item ID, keeping state local to each stream so parallel tool calls stay separate. A missing identity is rejected with a provider output error of kind invalid-tool-call, and a conflicting name is rejected rather than overwritten.

Completed tool calls remain validated by the shared parser. Final-only or encrypted reasoning that arrives only in the terminal Responses event, including on Azure streams, is retained by Core for later turns.

Provider identities

Surface	Identity
Completion, embedding, media handles	azure-openai
Chat Completions request trace	azure-openai-chat
Responses request trace	azure-openai-responses
Model listing errors	azure-openai

Not included

The package does not implement AI SDK-style Azure file-ID prefix inference, hosted-tool helpers, Azure Speech or MAI adapters, deployment management, or automatic credential discovery from environment variables.