Skip to main content

LangChain

There is no separate langchain-openproxyai package — because OpenProxyAI is OpenAI-API-compatible, LangChain's existing OpenAI integration works against it directly by overriding the base URL.

Python​

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="gpt-4o",
    base_url="https://api.openproxy.ai/v1",
    api_key=os.environ["OPENPROXY_KEY"],
    default_headers={
        "x-op-team": "eng-platform",
        "x-op-policy": "pii_redact,topic_guard",
    },
)

response = llm.invoke("Summarize this in one sentence.")

What you get, and what you don't​

Requests routed this way still go through the full request pipeline — policy enforcement, caching, budgets, and audit logging all apply exactly as they would for any other client. What you don't get through LangChain's standard ChatOpenAI wrapper is the typed gateway metadata (cost, latency, policy decision) that the dedicated Python SDK attaches to every response — that metadata is still present on the raw HTTP response headers, but LangChain's wrapper doesn't surface it by default.

Same pattern for JS/TS​

@langchain/openai's ChatOpenAI accepts the same baseURL override, with defaultHeaders for the OpenProxyAI-specific headers.