HFAI Cloud · integration
LangChain
on HFAI Cloud.
base_url and api_key on ChatOpenAI.
A chat model with two overrides.
Its own class,
its own two arguments.
LangChain documents its OpenAI integration around a chat-model class called ChatOpenAI, and that class takes base_url and api_key as arguments of its own. The same page notes that an older openai_api_base spelling is still accepted for the first of the two, and that the environment can supply them when the constructor does not. Checked against python.langchain.com/docs/integrations/chat/openai.
from langchain_openai import ChatOpenAI
model = ChatOpenAI(
model="gpt-oss:120b",
base_url="https://api.heartflowai.net/v1",
api_key="hfai_live_…",
)
print(model.invoke("Hello HFAI").content)
The instance above is the object a chain or an agent is handed, which means the endpoint decision is made once and everything built on the model inherits it. Model identifiers stay plain strings, so a different row of the catalogue is a different string and nothing else.
LangChain is an independent framework project with its own maintainers and its own release process. HeartFlowAI is not affiliated with it, is not a partner of it, and this page is not approved or endorsed by its maintainers. The two argument names and the import above were checked at python.langchain.com/docs/integrations/chat/openai.
The class a chain is built on.
Point it once,
build on top of it.
A chain, a tool-calling agent and a retriever all sit on top of the same chat-model object, so one construction decides the endpoint for the whole graph.
The chat-completions shape is in the API reference, the balance model on HFAI Cloud, and GPT-OSS 120B is the row the free tier covers.
Framework, not lock-in.
Because the class takes the endpoint as an argument, a LangChain application can keep its provider layer replaceable with one construction.