YouTipLangchain Deepseek
This chapter introduces how to integrate and use the DeepSeek chat model in LangChain.
Through the **langchain-deepseek** extension package, developers can quickly access the large language model services provided by DeepSeek.
DeepSeek models support both official hosted APIs and local or third-party inference deployment through platforms like Ollama, Fireworks, and Together.
> In addition to the official API interface, we can also directly access mainstream large models such as DeepSeek, Kimi, GLM, Doubao, and MiniMax through the [Coding Plan/Token Plan](#) package, without purchasing each vendor's API separately.
### DeepSeek Introduction
DeepSeek is an open-source large language model series that supports chat, reasoning, code generation, and other capabilities.
In LangChain, DeepSeek is primarily called through the **ChatDeepSeek** class.
Integration Information:
| Project | Description |
| --- | --- |
| Class Name | ChatDeepSeek |
| Installation Package | langchain-deepseek |
| Status | Beta |
| Supports JavaScript | Yes |
| Supports Python | Yes |
Model Capability Support:
| Feature | Supported |
| --- | --- |
| Tool Calling | ✅ Supported |
| Structured Output | ✅ Supported |
| Image Input | ❌ Not Supported |
| Audio Input | ❌ Not Supported |
| Video Input | ❌ Not Supported |
| Token Streaming Output | ✅ Supported |
| Native Async Call | ✅ Supported |
| Token Usage Statistics | ✅ Supported |
| Logprobs | ❌ Not Supported |
DeepSeek API uses OpenAI/Anthropic compatible API format. By modifying the configuration, you can use OpenAI/Anthropic SDK to access DeepSeek API, or use software compatible with OpenAI/Anthropic API.
| Parameter | Value |
| --- | --- |
| base_url (OpenAI) | `https://api.deepseek.com` |
| base_url (Anthropic) | `https://api.deepseek.com/anthropic` |
| api_key | Click the link to apply for (https://platform.deepseek.com/api_keys) |
| model* | `deepseek-v4-flash` `deepseek-v4-pro` `deepseek-chat` (will be deprecated on 2026/07/24) `deepseek-reasoner` (will be deprecated on 2026/07/24) |
* * *
## Install langchain-deepseek
Before starting to use, you need to install the DeepSeek LangChain integration package:
pip install -qU langchain-deepseek
* * *
## Configure DeepSeek API Key
You need to register a DeepSeek account first and create an API Key. If you don't have one yet, go to [https://platform.deepseek.com/api_keys](https://platform.deepseek.com/api_keys) to create an API key.
It is recommended to use **python-dotenv** to read the .env configuration file in the current directory.
pip install python-dotenv
Create a **.env** file in the current project directory:
## .env File Configuration:
DEEPSEEK_API_KEY="sk-xxxxxxxxxxxxxxxx"
OPENAI_API_KEY="sk-xxxxxxxxxxxxxxxx"
!(#)
Python reads .env file:
## Complete Example
import os
from dotenv import load_dotenv
# Load .env file in current directory
load_dotenv()
# Get API Key
api_key =os.getenv("DEEPSEEK_API_KEY")
print(api_key)
**Specify .env file path:** If the .env file is not in the current directory, you can manually specify the path:
## Example
from dotenv import load_dotenv
load_dotenv(dotenv_path="./config/.env")
### Configure LangSmith (Optional)
If you need to enable LangChain call chain tracking and debugging features, you can configure LangSmith API Key:
## Example
os.environ="true"
os.environ=getpass.getpass(
"Enter your LangSmith API key: "
)
### Create ChatDeepSeek Model
After installation, you can initialize the model through **ChatDeepSeek**:
## Example
import os
from dotenv import load_dotenv
from langchain_deepseek import ChatDeepSeek
# Load .env
load_dotenv()
# Get API KEY
api_key =os.getenv("DEEPSEEK_API_KEY")
# Create model
llm = ChatDeepSeek(
api_key=api_key,
model="deepseek-v4-flash",
temperature=0,
max_tokens=None,
timeout=None,
max_retries=2
)
# Call model
response = llm.invoke("Hello, please introduce LangChain")
print(response.content)
Parameter Description:
| Parameter | Description |
| --- | --- |
| api_key | Set your applied API key, e.g., "sk-xxx" |
| model | Specify model name, e.g., deepseek-v4-flash |
| temperature | Control randomness, lower values result in more stable outputs |
| max_tokens | Limit the maximum number of Tokens to generate |
| timeout | Request timeout |
| max_retries | Maximum retry attempts after failure |
You can also use the [init_chat_model()](#) function to call it:
## Example
import os
from dotenv import load_dotenv
# Load .env file in current directory
load_dotenv()
from langchain.chat_models import init_chat_model
# Specify model, returns fixed model
model = init_chat_model("deepseek:deepseek-v4-flash", temperature=0.7)
response = model.invoke("Introduce TUTORIAL")
print(response.content)
* * *
## Call DeepSeek Model
The following example demonstrates how to send chat messages to the DeepSeek model:
> Note: I wrote the API Key in the test file here for testing:
>
> llm = ChatDeepSeek( api_key="sk-xxx", # Set your DeepSeek API Key model="deepseek-v4-flash")
> In actual production environment, please set it in the .env file.
## Example
from langchain_deepseek import ChatDeepSeek
llm = ChatDeepSeek(
api_key="sk-xxx",# Set your DeepSeek API Key
model="deepseek-v4-flash",
temperature=0,
max_tokens=None,
timeout=None,
max_retries=2
)
messages =[
(
"system",
"You are a helpful assistant that translates English to Chinese."
),
(
"human",
"I love programming."
),
]
ai_msg = llm.invoke(messages)
print(ai_msg.content)
After running, the model will return the French translation result.
I love programming.
Supported DeepSeek Models:
| Model | Usage | Features |
| --- | --- | --- |
| deepseek-v4-flash | General chat model | Supports Tool Calling and structured output |
| deepseek-v4-pro | Reasoning model (deepseek-v4-pro) | Stronger reasoning capability, but does not support Tool Calling |
## LangChain + DeepSeek Complete Test Example
Below demonstrates a complete runnable LangChain + DeepSeek example.
### Create Test File
Create test.py file:
## Complete Example
from langchain_deepseek import ChatDeepSeek
from langchain_core.messages import HumanMessage, SystemMessage
# =========================
# Configure DeepSeek API Key
# =========================
apiKey ="sk-xxx"# Set your DeepSeek API Key
# =========================
# Create DeepSeek Model
# =========================
llm = ChatDeepSeek(
api_key=apiKey,
model="deepseek-v4-flash",
temperature=0.7,
max_tokens=1024,
timeout=60,
max_retries=3
)
# =========================
# Construct Chat Messages
# =========================
messages =[
SystemMessage(
content="You are a professional Python instructor."
),
HumanMessage(
content="Please explain what LangChain is and provide a simple example."
)
]
# =========================
# Call Model
# =========================
response = llm.invoke(messages)
# =========================
# Output Result
# =========================
print("AI Reply:")
print(response.content)
Execute in terminal:
python test.py
Output is as follows:
!(#)
Code Description:
| Code | Function |
| --- | --- |
| ChatDeepSeek | DeepSeek chat model class in LangChain |
| SystemMessage | System prompt, used to set AI identity |
| HumanMessage | User input message |
| llm.invoke() | Call DeepSeek model |
| response.content | Get AI returned content |
### Use deepseek-v4-pro Reasoning Model
LangChain supports DeepSeek streaming output, you need to use the deepseek-v4-pro reasoning model