Using DeepSeek with LangChain

Setting Up DeepSeek with LangChain

1. Create a DeepSeek Account and Get an API Key**

- Visit DeepSeek's website to create an account.
- Generate an API key, which is necessary for accessing DeepSeek models.

2. Install LangChain DeepSeek Integration Package**

You need to install the `langchain-deepseek` package to integrate DeepSeek models with LangChain. This can be done using pip:

bash
%pip install -qU langchain-deepseek

3. Set Environment Variables**

Set the `DEEPSEEK_API_KEY` environment variable with your API key:

python
import os
if not os.getenv("DEEPSEEK_API_KEY"):
    os.environ["DEEPSEEK_API_KEY"] = "your-api-key"

4. Instantiate DeepSeek Model in LangChain**

You can instantiate a DeepSeek model using the `ChatDeepSeek` class. For example, to use the `deepseek-chat` model:

python
from langchain_deepseek import ChatDeepSeek

llm = ChatDeepSeek(
    model="deepseek-chat",
    temperature=0,
    max_tokens=None,
    timeout=None,
    max_retries=2,
    # other params...
)

5. Invoke the Model**

To generate chat completions, you can invoke the model with a prompt:

python
messages = [
    ("system", "You are a helpful assistant that translates English to French. Translate the user sentence."),
    ("human", "I love programming."),
]

ai_msg = llm.invoke(messages)
print(ai_msg.content)

Features and Limitations

- Tool Calling: Supported by DeepSeek-V3 (`deepseek-chat`), but not by DeepSeek-R1 (`deepseek-reasoner`)[2][4].
- Structured Output: Also supported by DeepSeek-V3, not by DeepSeek-R1[2][4].
- Local Deployment: DeepSeek models can be run locally using providers like Ollama[2][4].

Building Applications with DeepSeek and LangChain

You can build advanced applications like Retrieval-Augmented Generation (RAG) systems using DeepSeek, LangChain, and tools like Streamlit for interactive interfaces[3][7]. These systems allow you to query local documents and answer complex questions.