@ank1015/agents-provider-openai-azure

Azure OpenAI Responses API runtime adapter for @ank1015/agents.

This package turns normalized LLMRequest<"openai-azure"> objects into Azure OpenAI Responses API streaming calls and translates response stream events back into normalized assistant events and messages. It also re-exports @ank1015/agents-provider-openai-azure-spec from the root entrypoint.

Install

pnpm add @ank1015/agents-provider-openai-azure

Exports

• @ank1015/agents-provider-openai-azure
• @ank1015/agents-provider-openai-azure/adapter

Usage

import { createAdapterTransport } from '@ank1015/agents-core/transport';
import {
  OPENAI_AZURE_PROVIDER,
  createOpenAIAzureProviderAdapter,
} from '@ank1015/agents-provider-openai-azure';

const transport = createAdapterTransport([
  createOpenAIAzureProviderAdapter({
    provider: OPENAI_AZURE_PROVIDER,
    apiKey: { type: 'env', name: 'AZURE_OPENAI_API_KEY' },
    endpoint: 'https://example.openai.azure.com',
  }),
]);

When the key comes from another secret source, pass it directly:

createOpenAIAzureProviderAdapter({
  provider: OPENAI_AZURE_PROVIDER,
  apiKey: { type: 'value', value: azureOpenAIKey },
  endpoint: 'https://example.openai.azure.com',
});

When apiVersion is omitted, the adapter uses Azure OpenAI v1/OpenAI-compatible URLs. When apiVersion is present, it creates the legacy AzureOpenAI SDK client.

The wire request model is always request.modelId; Azure deployment names are expected to match the static catalog model ids.

The catalog includes GPT-5.6 Sol, Terra, and Luna. GPT-5.6 request options such as max reasoning effort, Pro mode, persisted reasoning context, and request-wide prompt-cache options pass through providerOptions. Usage reports cache reads and cache writes separately. For GPT-5.6 and GPT-5.5, the adapter selects whole-request long-context rates when total input exceeds 272K tokens.