LangChain integrations

Available in Classic and VPC

You can use LangChain to easily integrate and use CLOVA Studio's HyperCLOVA X model and embedding tools.

LangChain is an open source framework that supports language model-based applications. You can link various tools, including language models like HyperCLOVA X, vector databases, and search engines, in a chain structure, simplifying the process of connecting features and integrating the development process. LangGraph, an AI agent development framework, can also be utilized through LangChain integrations.

This LangChain integrations guide describes how to install LangChain and configure settings for integrating CLOVA Studio. Additionally, it provides example codes for utilizing CLOVA Studio's HyperCLOVA X model and embedding tools through LangChain for practical references for real world development.

Install and verify LangChain

To use LangChain integrated with CLOVA Studio, Python version 3.9 or higher is required. After you install Python, use the following commands to install LangChain, then install the langchain-naver package necessary for integrations.

pip install -qU langchain # install LangChain
pip install -qU langchain-naver # install LangChain-NAVER integration package

Verify scope of integrations

CLOVA Studio offers these key features you can use through LangChain:

• Chat mode model in CLOVA Studio's Playground
  • Model: Basic model (Example: HCX-005) and tuned models of the basic model
    • Associated APIs: Chat completions, chat completions v3, and OpenAI compatibility
  • Integration examples: Use HyperCLOVA X model
• Embedding (or embedding v2) tool in CLOVA Studio's Explorer
  • Model: clir-emb-dolphin, clir-sts-dolphin, and bge-m3
    • Associated APIs: Embedding, embedding v2, and OpenAI compatibility
  • Integration example: Use embedding tools

Integration settings

To securely use CLOVA Studio features through LangChain, register the API key as an environment variable, which acts as the authentication information required for API calls. You can issue and check your API key from the [API key] menu in CLOVA Studio. Create the service app when you apply it to an actual service.

To register the authentication information as environment variables:

• Registration through the terminal

  export CLOVASTUDIO_API_KEY="<CLOVA-STUDIO-API-KEY>"
  

• Registration through Python

  import getpass
  import os
  
  os.environ["CLOVASTUDIO_API_KEY"] = getpass.getpass(
          "Enter CLOVA Studio API key: "
      )
  

Integration examples

This section introduces example codes for integrating LangChain with CLOVA Studio.

• Example using CLOVA Studio's HyperCLOVA X model through LangChain
• Example using CLOVA Studio's embedding tools through LangChain

Use HyperCLOVA X model

The following is an example code for using CLOVA Studio's HyperCLOVA X model through LangChain:

from langchain_naver import ChatClovaX
  
chat = ChatClovaX(
    model="HCX-005" # Enter model name (default value: HCX-005) 
)

You can use the following parameters with your request when defining the ChatClovaX class instance:

You can run chat models using invoke or stream, LangChain's Chat model class methods. The following are example codes:

• invoke

  messages = [
      (
          "system",
          "CLOVA Studio is a development tool that allows you to easily create AI services using the HyperCLOVA X model.",
      ),
      ("human", "What is CLOVA Studio?"),
  ]
  ai_msg = chat.invoke(messages)
  ai_msg
  

• stream

  messages = [
      (
          "system",
          "CLOVA Studio is a development tool that allows you to easily create AI services using the HyperCLOVA X model.",
      ),
      ("human", "What is CLOVA Studio?"),
  ]
  
  for chunk in chat.stream(messages):
      print(chunk.content, end="", flush=True)
  

Additionally, you can use the bind_tools method to implement LangChain's tool calling feature based on CLOVA Studio's feature calling feature. The following are example codes:

from langchain_naver import ChatClovaX
from pydantic import BaseModel, Field

chat = ChatClovaX(
    model="HCX-005" # Enter the model name (default value: HCX-005)
    max_tokens=1024 # Set the value to more than 1024 when using Function calling
)

class GetWeather(BaseModel):
    '''Check the current weather for the given location.'''

    location: str = Field(
        ..., description="The name of the city where you want to check the weather. Example: Gyeonggi-do, Seongnam-si, Bundang-gu"
    )

chat_with_tool = chat.bind_tools(
    [GetWeather]
)
ai_msg = chat_with_tools.invoke(
    "Is it hotter in Bundang or Pangyo?"
)
ai_msg.tool_calls

If you use the thinking model, you can set related features through the thinking parameter when you define the ChatClovaX class instance. You can also check the reasoning process using the following separate method.

from langchain_naver import ChatClovaX

chat = ChatClovaX(
    model="HCX-007",
    thinking={
        "effort": "low"  # 'none' (do not use inference), 'low' (default value), 'medium', or 'high'
    },
)
ai_msg = chat.invoke("What is 3 cubed?")

print(ai_msg.content) # Final response
print(ai_msg.additional_kwargs["thinking_content"]) # Reasoning process

Use embedding tools

The following is an example code for using CLOVA Studio's embedding (embedding v2) tool through LangChain:

from langchain_naver import ClovaXEmbeddings
 
embeddings = ClovaXEmbeddings(
    model="clir-emb-dolphin",  # Model name (default value: clir-emb-dolphin)
)

You can use the following parameters with your request when defining the ClovaXEmbeddings class instance:

You can run chat models using embed_query orembed_documents, LangChain's Embedding model class methods. The following are example codes:

• embed_query

  query = "CLOVA Studio is a development tool that allows you to easily create AI services using the HyperCLOVA X model."
  single_vector = embeddings.embed_query(query)
  

• embed_documents

  text1="CLOVA Studio is a development tool that allows you to easily create AI services using the HyperCLOVA X model."
  text2 = "LangChain is an open source framework designed to support the development of language model-based applications."
  document = [text1, text2]
  multiple_vector = embeddings.embed_documents(document)