lmux-openai

OpenAI provider for lmux. Talks to the OpenAI REST API directly over 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.

Auth

Set OPENAI_API_KEY in your environment. The default OpenAIEnvAuthProvider reads it automatically.

from lmux_openai import OpenAIProvider

provider = OpenAIProvider()

Or pass a custom auth provider:

provider = OpenAIProvider(auth=my_auth_provider)

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

response = provider.create_response("gpt-4o", "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("openai", provider)
response = registry.chat("openai/gpt-4o", messages)

Provider Params

Pass OpenAI-specific parameters via provider_params:

from lmux_openai import OpenAIParams

response = provider.chat(
    "o3",
    messages,
    provider_params=OpenAIParams(reasoning_effort="high", service_tier="flex"),
)

Constructor Options

OpenAIProvider(
    auth=...,             # AuthProvider[str], default: OpenAIEnvAuthProvider()
    base_url=...,         # Optional base URL override
    timeout=...,          # Request timeout in seconds
    max_retries=...,      # Max retry attempts
    data_residency=...,   # bool, default: False — apply 10% uplift for regional endpoints
    organization=...,     # Optional org id -> OpenAI-Organization header
    project=...,          # Optional project id -> OpenAI-Project header
    default_headers=...,  # Optional Mapping[str, str] added to every request
    transport=...,        # Optional httpx.BaseTransport for the sync client (proxies, testing)
    async_transport=...,  # Optional httpx.AsyncBaseTransport for the async client
)

lmux does not read OpenAI's OPENAI_BASE_URL / OPENAI_ORG_ID / OPENAI_PROJECT_ID environment variables (only the API key, via OpenAIEnvAuthProvider). Pass base_url, organization, and project explicitly instead.

Custom Headers

default_headers applies to every request — useful for gateways and proxies (e.g. a Helicone-Auth token). lmux-managed headers (Authorization, Content-Type, OpenAI-Organization, OpenAI-Project) take precedence and cannot be overridden by default_headers; use organization / project for those.

provider = OpenAIProvider(
    organization="org-abc",
    project="proj-123",
    default_headers={"Helicone-Auth": "Bearer sk-helicone-..."},
)

Data Residency

OpenAI charges a 10% uplift on the gpt-5.4, gpt-5.5, and gpt-5.6 families when requests go through a regional processing (data residency) endpoint.

Data residency is selected at the transport layer (regional hostname like eu.api.openai.com), not via a per-request parameter. Set data_residency=True on the provider so lmux applies the uplift to the reported cost.

provider = OpenAIProvider(
    base_url="https://eu.api.openai.com/v1",
    data_residency=True,
)

The uplift is only applied to eligible models (checked via regional_uplift_applies); other models (e.g. gpt-4o, embeddings) return their standard cost even when data_residency=True.