langchain-runcrate

LangChain chat model for Runcrate: 200+ open models (DeepSeek, Qwen, Llama, Kimi, gpt-oss and more) behind one OpenAI-compatible API, billed per token from one prepaid balance. The same account rents GPUs by the hour when you want to fine-tune or serve your own model.

Install

pip install -U langchain-runcrate

Create an API key (docs) and export it:

export RUNCRATE_API_KEY="..."

Chat

from langchain_runcrate import ChatRuncrate

llm = ChatRuncrate(model="deepseek-ai/DeepSeek-V3.2", temperature=0)
print(llm.invoke("Explain LoRA fine-tuning in one sentence.").content)

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

Tool calling and structured output

from pydantic import BaseModel
from langchain_core.tools import tool


@tool
def gpu_price(gpu: str) -> str:
    """Return the hourly price of a GPU."""
    return "see https://www.runcrate.ai/api/gpu-pricing"


llm_with_tools = llm.bind_tools([gpu_price])
print(llm_with_tools.invoke("What does an H100 cost per hour?").tool_calls)


class Summary(BaseModel):
    title: str
    bullets: list[str]


print(llm.with_structured_output(Summary).invoke("Summarize what a GPU cloud is."))

Agents (LangGraph / create_agent)

from langchain.agents import create_agent

agent = create_agent(ChatRuncrate(model="Qwen/Qwen3-235B-A22B-Instruct-2507"), tools=[gpu_price])
agent.invoke({"messages": [{"role": "user", "content": "Price an H100 for 10 hours."}]})

Reasoning models

Reasoning models such as deepseek-ai/DeepSeek-R1-0528 return their reasoning in message.additional_kwargs["reasoning_content"], both in invoke and in stream.

Configuration

Every other ChatOpenAI argument (temperature, max_tokens, timeout, max_retries, streaming, model_kwargs ...) works the same way.

Models and per-token prices: runcrate.ai/inference/pricing.

Let your agent rent GPUs too

Runcrate also runs a remote MCP server at https://mcp.runcrate.ai/mcp (OAuth sign-in, no key) so Claude Code, Cursor or any MCP client can launch and drive GPU instances. Setup: runcrate.ai/docs/mcp/setup, client packages: aeonmindai/runcrate-mcp.

Development

uv run --group test pytest tests/unit_tests
RUNCRATE_API_KEY=... uv run --group test pytest tests/integration_tests

License

MIT, Aeonmind LLC.