RunPod provides GPU cloud infrastructure, including Serverless endpoints optimized for deploying and scaling AI models. This guide covers how to use the langchain-runpod integration package to connect LangChain applications to models hosted on RunPod Serverless. The integration offers interfaces for both standard Language Models (LLMs) and Chat Models.

​
Intstallation

Install the dedicated partner package: 
%pip install -qU langchain-runpod

​
Setup

​
1. Deploy an Endpoint on RunPod

  • Navigate to your RunPod Serverless Console.
  • Create a “New Endpoint”, selecting an appropriate GPU and template (e.g., vLLM, TGI, text-generation-webui) compatible with your model and the expected input/output format (see component guides or the package README).
  • Configure settings and deploy.
  • Crucially, copy the Endpoint ID after deployment.

​
2. Set API Credentials

The integration needs your RunPod API Key and the Endpoint ID. Set them as environment variables for secure access: 
import getpass
import os

os.environ["RUNPOD_API_KEY"] = getpass.getpass("Enter your RunPod API Key: ")
os.environ["RUNPOD_ENDPOINT_ID"] = input("Enter your RunPod Endpoint ID: ")
(Optional) If using different endpoints for LLM and Chat models, you might need to set RUNPOD_CHAT_ENDPOINT_ID or pass the ID directly during initialization.

​
Components

This package provides two main components:

​
1. LLM

For interacting with standard text completion models. See the RunPod LLM Integration Guide for detailed usage 
from langchain_runpod import RunPod

# Example initialization (uses environment variables)
llm = RunPod(model_kwargs={"max_new_tokens": 100})  # Add generation params here

# Example Invocation
try:
    response = llm.invoke("Write a short poem about the cloud.")
    print(response)
except Exception as e:
    print(
        f"Error invoking LLM: {e}. Ensure endpoint ID and API key are correct and endpoint is active."
    )

​
2. Chat Model

For interacting with conversational models. See the RunPod Chat Model Integration Guide for detailed usage and feature support. 
from langchain_core.messages import HumanMessage
from langchain_runpod import ChatRunPod

# Example initialization (uses environment variables)
chat = ChatRunPod(model_kwargs={"temperature": 0.8})  # Add generation params here

# Example Invocation
try:
    response = chat.invoke(
        [HumanMessage(content="Explain RunPod Serverless in one sentence.")]
    )
    print(response.content)
except Exception as e:
    print(
        f"Error invoking Chat Model: {e}. Ensure endpoint ID and API key are correct and endpoint is active."
    )