lmux-azure-foundry

Azure AI Foundry provider for lmux. Talks to the Azure OpenAI REST API directly with httpx.

Supports chat completions, streaming, embeddings, and the Responses API.

Part of the lmux ecosystem: standardized interface, cost tracking on every response, and registry-based routing across providers.

Optional Extras

• lmux-azure-foundry[identity]: Azure AD token authentication via azure-identity

Auth

Three authentication methods:

API Key (default)

Set AZURE_FOUNDRY_API_KEY in your environment:

from lmux_azure_foundry import AzureFoundryProvider

provider = AzureFoundryProvider(endpoint="https://your-resource.openai.azure.com")

Azure AD Token

from lmux_azure_foundry import AzureFoundryProvider, AzureAdToken

provider = AzureFoundryProvider(
    endpoint="https://your-resource.openai.azure.com",
    auth=my_auth_returning_azure_ad_token,
)

Token Provider

from lmux_azure_foundry import AzureFoundryTokenAuthProvider

provider = AzureFoundryProvider(
    endpoint="https://your-resource.openai.azure.com",
    auth=AzureFoundryTokenAuthProvider(),  # uses azure-identity DefaultAzureCredential
)

Usage

Chat

from lmux import UserMessage

response = provider.chat("gpt-4o", [UserMessage(content="Hello")])
print(response.content)
print(response.cost)

Streaming

for chunk in provider.chat_stream("gpt-4o", [UserMessage(content="Hello")]):
    if chunk.delta:
        print(chunk.delta, end="")

Embeddings

response = provider.embed("text-embedding-3-small", "Hello")
print(response.embeddings)

Responses API

Required for models that are only served through the Responses API:

response = provider.create_response("gpt-5-pro", "Hello")
print(response.output_text)

Async

All methods have async variants: achat, achat_stream, aembed, acreate_response.

Registry

Use with the lmux registry to route across multiple providers:

from lmux import Registry

registry = Registry()
registry.register("azure", provider)
response = registry.chat("azure/gpt-4o", messages)

Provider Params

from lmux_azure_foundry import AzureFoundryParams

response = provider.chat(
    "gpt-4o",
    messages,
    provider_params=AzureFoundryParams(deployment_type="data_zone"),
)

deployment_type="data_zone" uses US pricing when data_zone is omitted. Set data_zone="eu" for an EU Data Zone deployment. This selects the applicable model-specific premium for cost reporting; it does not change API routing. Other deployment types ignore data_zone.

Constructor Options

AzureFoundryProvider(
    endpoint=...,      # required, Azure resource endpoint
    auth=...,          # AuthProvider, default: AzureFoundryKeyAuthProvider()
    api_version=...,   # API version (default: "2025-04-01-preview")
    timeout=...,       # Request timeout in seconds
    max_retries=...,   # Max retry attempts
    default_headers=...,  # Optional headers included with every request
    transport=...,        # Optional httpx.BaseTransport for the sync client (proxies, testing)
    async_transport=...,  # Optional httpx.AsyncBaseTransport for the async client
)

default_headers is useful for gateway authentication, tracing, and routing. Foundry-managed authentication and content-type headers take precedence over caller values, case-insensitively.