# LangChain

> Point langchain-openai's base_url at TokRouter and use any gateway model inside LangChain.

LangChain's OpenAI integration accepts a custom `base_url` — point it at the gateway.

```bash
pip install langchain-openai
```

```python
import os
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="gpt-5.4-mini",
    api_key=os.environ["TOKROUTER_API_KEY"],
    base_url="https://api.tokrouter.dev/v1",
)

print(llm.invoke("Explain RAG in one sentence").content)
```

## Mixing vendors in one chain

```python
planner = ChatOpenAI(
    model="claude-opus-5-5",
    api_key=os.environ["TOKROUTER_API_KEY"],
    base_url="https://api.tokrouter.dev/v1",
)
worker = ChatOpenAI(
    model="deepseek-v4-flash",
    api_key=os.environ["TOKROUTER_API_KEY"],
    base_url="https://api.tokrouter.dev/v1",
)

plan = planner.invoke("Split 'refactor the login module' into three steps; output only the list").content
print(worker.invoke(f"Write the implementation notes for step one:\n{plan}").content)
```

<Callout type="info" title="Streaming and tools">
`stream()` and `bind_tools()` use the same OpenAI protocol fields, and the gateway behaves identically.
Tool-use support varies by model — see [Tool calling](/en/guides/tool-calling/).
</Callout>