Anvia

@anvia/azure API reference ​

The package exports AzureOpenAIClient, its AzureOpenAIClientOptions type, and AzureOpenAI-prefixed aliases for the shared OpenAI completion, embedding, image generation, speech generation, and transcription option and handle types. It has no public subpath exports.

Client ​

ts
import type OpenAI from 'openai'

type AzureOpenAIClientOptions =
  | ({ endpoint: string; baseUrl?: never } | { baseUrl: string; endpoint?: never }) &
      (
        | { apiKey: string; azureADTokenProvider?: never }
        | { azureADTokenProvider: () => Promise<string>; apiKey?: never }
      ) & {
        headers?: Record<string, string>
        fetch?: typeof fetch
        client?: never
      }
  | {
      client: OpenAI
      endpoint?: never
      baseUrl?: never
      apiKey?: never
      azureADTokenProvider?: never
      headers?: never
      fetch?: never
    }

class AzureOpenAIClient {
  constructor(options: AzureOpenAIClientOptions)

  completionModel(options: AzureOpenAICompletionModelOptions): AzureOpenAICompletionModel
  embeddingModel(options: AzureOpenAIEmbeddingModelOptions): AzureOpenAIEmbeddingModelHandle
  imageGenerationModel(
    options: AzureOpenAIImageGenerationModelOptions,
  ): AzureOpenAIImageGenerationModelHandle
  speechGenerationModel(
    options: AzureOpenAISpeechGenerationModelOptions,
  ): AzureOpenAISpeechGenerationModelHandle
  transcriptionModel(
    options: AzureOpenAITranscriptionModelOptions,
  ): AzureOpenAITranscriptionModelHandle
  listModels(options?: { abortSignal?: AbortSignal }): Promise<ModelList>
}

completionModel({ modelId, api?, contextLimits?, controls? }) takes the deployment name as modelId, with api defaulting to 'chat'. The option and handle types are the corresponding @anvia/openai types under an AzureOpenAI prefix:

Azure aliasSource type
AzureOpenAICompletionModel, AzureOpenAICompletionModelOptionsOpenAICompletionModel, OpenAICompletionModelOptions
AzureOpenAIEmbeddingModelHandle, AzureOpenAIEmbeddingModelOptionsOpenAIEmbeddingModelHandle, OpenAIEmbeddingModelOptions
AzureOpenAIImageGenerationModelHandle, AzureOpenAIImageGenerationModelOptionsOpenAIImageGenerationModelHandle, OpenAIImageGenerationModelOptions
AzureOpenAISpeechGenerationModelHandle, AzureOpenAISpeechGenerationModelOptionsOpenAISpeechGenerationModelHandle, OpenAISpeechGenerationModelOptions
AzureOpenAITranscriptionModelHandle, AzureOpenAITranscriptionModelOptionsOpenAITranscriptionModelHandle, OpenAITranscriptionModelOptions

Handles report provider: 'azure-openai'. Chat request traces report azure-openai-chat and Responses traces report azure-openai-responses. listModels() failures throw a ModelListingError with provider: 'azure-openai' and the original error as cause.

Errors ​

Construction throws TypeError for invalid endpoint, credential, or option combinations, as listed in Configuration. Azure Responses streams that cannot resolve or that conflict on a function name reject with a CompletionProviderOutputError of kind invalid-tool-call.

@anvia/openai/adapters (added in @anvia/openai 1.2.0) exports OpenAIResponsesCompletionModel, the reusable Responses model that provider packages extend. See Migrate from @anvia/openai.

Return to the @anvia/azure overview.

Built for Anvia.