langchain-outrun

LangChain integration for Outrun, a prepaid, OpenAI-compatible API. One Outrun key reaches GPT, Claude, Gemini, Grok, DeepSeek, GLM, Kimi, Qwen and MiniMax models.

ChatOutrun is a LangChain chat model built on langchain-openai, with the Outrun base URL and key set for you.

Install

pip install -U langchain-outrun

Create a key at https://outrun.to/key, then:

export OUTRUN_API_KEY="your-api-key"

Chat model

from langchain_outrun import ChatOutrun

model = ChatOutrun(model="gpt-5.6-terra", temperature=0)
print(model.invoke("Explain prompt caching in one sentence.").content)

Pick any model id from https://outrun.to/pricing (for example claude-haiku-5-5, gemini-3.8-flash or deepseek-v4.1-flash). The same list is available as JSON, without a key, at https://api.outrun.to/pricing.

Streaming

for chunk in model.stream("Write a haiku about rain."):
    print(chunk.text, end="", flush=True)

Token usage is included in streamed responses (usage_metadata on the last chunk).

Tool calling

from pydantic import BaseModel, Field


class GetWeather(BaseModel):
    """Get the current weather for a city."""

    city: str = Field(description="City name")


model_with_tools = model.bind_tools([GetWeather])
msg = model_with_tools.invoke("What's the weather in Paris?")
print(msg.tool_calls)

Structured output

class Answer(BaseModel):
    answer: str
    confidence: float


structured = model.with_structured_output(Answer)
print(structured.invoke("Is the sky blue? Answer briefly."))

Tool calling and structured output depend on the model you pick.

Async

ainvoke, astream and abatch work as with any LangChain chat model.

Settings

Other ChatOpenAI arguments such as temperature, max_tokens, timeout, max_retries and default_headers work the same way.

• Requests go to /v1/chat/completions. Pass use_responses_api=True to use /v1/responses instead (GPT models only).
• ChatOutrun never reads OPENAI_API_KEY, so an OpenAI key is never sent to Outrun.
• Each request carries an X-Outrun-SDK: langchain/<version> header.

Development

pip install -e . pytest pytest-asyncio pytest-socket "langchain-tests>=1,<2"
pytest tests/unit_tests --disable-socket --allow-unix-socket

tests/integration_tests call the live API (billed to your balance) and run only when OUTRUN_API_KEY is set.

Links

• Outrun docs: https://outrun.to/docs/
• Plain OpenAI SDK wrapper: outrun-ai
• Support: support@outrun.to

License: MIT.