LangGraph智能体开发快速入门(第2章)-从零构建LangGraph智能体

莫然

774人浏览 · 2025-08-18 08:27:33

莫然 · 2025-08-18 08:27:33 发布

二、从零构建LangGraph智能体

1. LangGraph接入大模型流程

LangGraph安装

本次公开课使用的版本是：langgraph==0.4.8

Python
! pip install langgraph -i https://pypi.tuna.tsinghua.edu.cn/simple

Plaintext
Collecting langgraph
Downloading langgraph-0.4.8-py3-none-any.whl.metadata (6.8 kB)
Requirement already satisfied: langchain-core>=0.1 in /root/anaconda3/lib/python3.12/site-packages (from langgraph) (0.3.65)
Collecting langgraph-checkpoint>=2.0.26 (from langgraph)
Downloading langgraph_checkpoint-2.1.0-py3-none-any.whl.metadata (4.2 kB)
Collecting langgraph-prebuilt>=0.2.0 (from langgraph)
Downloading langgraph_prebuilt-0.2.2-py3-none-any.whl.metadata (4.5 kB)
Collecting langgraph-sdk>=0.1.42 (from langgraph)
Downloading langgraph_sdk-0.1.70-py3-none-any.whl.metadata (1.5 kB)
Requirement already satisfied: pydantic>=2.7.4 in /root/anaconda3/lib/python3.12/site-packages (from langgraph) (2.11.4)
Requirement already satisfied: xxhash>=3.5.0 in /root/anaconda3/lib/python3.12/site-packages (from langgraph) (3.5.0)
Requirement already satisfied: langsmith<0.4,>=0.3.45 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core>=0.1->langgraph) (0.3.45)
Requirement already satisfied: tenacity!=8.4.0,<10.0.0,>=8.1.0 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core>=0.1->langgraph) (8.5.0)
Requirement already satisfied: jsonpatch<2.0,>=1.33 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core>=0.1->langgraph) (1.33)
Requirement already satisfied: PyYAML>=5.3 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core>=0.1->langgraph) (6.0.1)
Requirement already satisfied: packaging<25,>=23.2 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core>=0.1->langgraph) (24.1)
Requirement already satisfied: typing-extensions>=4.7 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core>=0.1->langgraph) (4.13.2)
Collecting ormsgpack>=1.10.0 (from langgraph-checkpoint>=2.0.26->langgraph)
Downloading ormsgpack-1.10.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (43 kB)
?[2K     ?[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━?[0m ?[32m43.7/43.7 kB?[0m ?[31m14.2 kB/s?[0m eta ?[36m0:00:00?[0ma ?[36m0:00:01?[0m
?[?25hRequirement already satisfied: httpx>=0.25.2 in /root/anaconda3/lib/python3.12/site-packages (from langgraph-sdk>=0.1.42->langgraph) (0.28.1)
Requirement already satisfied: orjson>=3.10.1 in /root/anaconda3/lib/python3.12/site-packages (from langgraph-sdk>=0.1.42->langgraph) (3.10.13)
Requirement already satisfied: annotated-types>=0.6.0 in /root/anaconda3/lib/python3.12/site-packages (from pydantic>=2.7.4->langgraph) (0.6.0)
Requirement already satisfied: pydantic-core==2.33.2 in /root/anaconda3/lib/python3.12/site-packages (from pydantic>=2.7.4->langgraph) (2.33.2)
Requirement already satisfied: typing-inspection>=0.4.0 in /root/anaconda3/lib/python3.12/site-packages (from pydantic>=2.7.4->langgraph) (0.4.0)
Requirement already satisfied: anyio in /root/anaconda3/lib/python3.12/site-packages (from httpx>=0.25.2->langgraph-sdk>=0.1.42->langgraph) (4.8.0)
Requirement already satisfied: certifi in /root/anaconda3/lib/python3.12/site-packages (from httpx>=0.25.2->langgraph-sdk>=0.1.42->langgraph) (2024.2.2)
Requirement already satisfied: httpcore==1.* in /root/anaconda3/lib/python3.12/site-packages (from httpx>=0.25.2->langgraph-sdk>=0.1.42->langgraph) (1.0.7)
Requirement already satisfied: idna in /root/anaconda3/lib/python3.12/site-packages (from httpx>=0.25.2->langgraph-sdk>=0.1.42->langgraph) (3.7)
Requirement already satisfied: h11<0.15,>=0.13 in /root/anaconda3/lib/python3.12/site-packages (from httpcore==1.*->httpx>=0.25.2->langgraph-sdk>=0.1.42->langgraph) (0.14.0)
Requirement already satisfied: jsonpointer>=1.9 in /root/anaconda3/lib/python3.12/site-packages (from jsonpatch<2.0,>=1.33->langchain-core>=0.1->langgraph) (2.1)
Requirement already satisfied: requests<3,>=2 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.3.45->langchain-core>=0.1->langgraph) (2.32.3)
Requirement already satisfied: requests-toolbelt<2.0.0,>=1.0.0 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.3.45->langchain-core>=0.1->langgraph) (1.0.0)
Requirement already satisfied: zstandard<0.24.0,>=0.23.0 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.3.45->langchain-core>=0.1->langgraph) (0.23.0)
Requirement already satisfied: charset-normalizer<4,>=2 in /root/anaconda3/lib/python3.12/site-packages (from requests<3,>=2->langsmith<0.4,>=0.3.45->langchain-core>=0.1->langgraph) (2.0.4)
Requirement already satisfied: urllib3<3,>=1.21.1 in /root/anaconda3/lib/python3.12/site-packages (from requests<3,>=2->langsmith<0.4,>=0.3.45->langchain-core>=0.1->langgraph) (2.2.2)
Requirement already satisfied: sniffio>=1.1 in /root/anaconda3/lib/python3.12/site-packages (from anyio->httpx>=0.25.2->langgraph-sdk>=0.1.42->langgraph) (1.3.0)
Downloading langgraph-0.4.8-py3-none-any.whl (152 kB)
?[2K   ?[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━?[0m ?[32m152.4/152.4 kB?[0m ?[31m6.7 kB/s?[0m eta ?[36m0:00:00?[0ma ?[36m0:00:02?[0m
?[?25hDownloading langgraph_checkpoint-2.1.0-py3-none-any.whl (43 kB)
?[2K   ?[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━?[0m ?[32m43.8/43.8 kB?[0m ?[31m26.7 kB/s?[0m eta ?[36m0:00:00?[0ma ?[36m0:00:01?[0m
?[?25hDownloading langgraph_prebuilt-0.2.2-py3-none-any.whl (23 kB)
Downloading langgraph_sdk-0.1.70-py3-none-any.whl (49 kB)
?[2K   ?[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━?[0m ?[32m50.0/50.0 kB?[0m ?[31m27.4 kB/s?[0m eta ?[36m0:00:00?[0ma ?[36m0:00:01?[0m
?[?25hDownloading ormsgpack-1.10.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (216 kB)
?[2K   ?[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━?[0m ?[32m216.7/216.7 kB?[0m ?[31m10.7 kB/s?[0m eta ?[36m0:00:00?[0m00:01?[0m00:02?[0m
?[?25hInstalling collected packages: ormsgpack, langgraph-sdk, langgraph-checkpoint, langgraph-prebuilt, langgraph
Successfully installed langgraph-0.4.8 langgraph-checkpoint-2.1.0 langgraph-prebuilt-0.2.2 langgraph-sdk-0.1.70 ormsgpack-1.10.0
?[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv?[0m?[33m
?[0m

Python
! pip show langgraph

Plaintext
Name: langgraph
Version: 0.4.8
Summary: Building stateful, multi-actor applications with LLMs
Home-page:
Author:
Author-email:
License:
Location: /root/anaconda3/lib/python3.12/site-packages
Requires: langchain-core, langgraph-checkpoint, langgraph-prebuilt, langgraph-sdk, pydantic, xxhash
Required-by:

LangGraph接入DeepSeek模型流程

接下来以DeepSeek为例，介绍LangGraph接入大模型流程。

Python
! pip install python-dotenv openai

Plaintext
Requirement already satisfied: python-dotenv in /root/anaconda3/lib/python3.12/site-packages (1.0.1)
Requirement already satisfied: openai in /root/anaconda3/lib/python3.12/site-packages (1.78.1)
Requirement already satisfied: anyio<5,>=3.5.0 in /root/anaconda3/lib/python3.12/site-packages (from openai) (4.8.0)
Requirement already satisfied: distro<2,>=1.7.0 in /root/anaconda3/lib/python3.12/site-packages (from openai) (1.9.0)
Requirement already satisfied: httpx<1,>=0.23.0 in /root/anaconda3/lib/python3.12/site-packages (from openai) (0.28.1)
Requirement already satisfied: jiter<1,>=0.4.0 in /root/anaconda3/lib/python3.12/site-packages (from openai) (0.8.2)
Requirement already satisfied: pydantic<3,>=1.9.0 in /root/anaconda3/lib/python3.12/site-packages (from openai) (2.11.4)
Requirement already satisfied: sniffio in /root/anaconda3/lib/python3.12/site-packages (from openai) (1.3.0)
Requirement already satisfied: tqdm>4 in /root/anaconda3/lib/python3.12/site-packages (from openai) (4.66.4)
Requirement already satisfied: typing-extensions<5,>=4.11 in /root/anaconda3/lib/python3.12/site-packages (from openai) (4.13.2)
Requirement already satisfied: idna>=2.8 in /root/anaconda3/lib/python3.12/site-packages (from anyio<5,>=3.5.0->openai) (3.7)
Requirement already satisfied: certifi in /root/anaconda3/lib/python3.12/site-packages (from httpx<1,>=0.23.0->openai) (2024.2.2)
Requirement already satisfied: httpcore==1.* in /root/anaconda3/lib/python3.12/site-packages (from httpx<1,>=0.23.0->openai) (1.0.7)
Requirement already satisfied: h11<0.15,>=0.13 in /root/anaconda3/lib/python3.12/site-packages (from httpcore==1.*->httpx<1,>=0.23.0->openai) (0.14.0)
Requirement already satisfied: annotated-types>=0.6.0 in /root/anaconda3/lib/python3.12/site-packages (from pydantic<3,>=1.9.0->openai) (0.6.0)
Requirement already satisfied: pydantic-core==2.33.2 in /root/anaconda3/lib/python3.12/site-packages (from pydantic<3,>=1.9.0->openai) (2.33.2)
Requirement already satisfied: typing-inspection>=0.4.0 in /root/anaconda3/lib/python3.12/site-packages (from pydantic<3,>=1.9.0->openai) (0.4.0)
?[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv?[0m?[33m
?[0m

这里我们使用DeepSeek官方的API_KEK进行调用。如果初次使用，需要现在DeepSeek官网上进行注册并创建一个新的API_Key，其官方地址为：https://platform.deepseek.com/usage

注册好DeepSeek的API_KEY后，首先在项目同级目录下创建一个env文件，用于存储DeepSeek的API_KEY，如下所示：

接下来通过python-dotenv库读取env文件中的API_KEY，使其加载到当前的运行环境中，代码如下：

Python
import os
from dotenv import load_dotenv
load_dotenv(override=True)

DeepSeek_API_KEY = os.getenv("DEEPSEEK_API_KEY")
# print(DeepSeek_API_KEY) # 可以通过打印查看

Python
from openai import OpenAI

# 初始化DeepSeek的API客户端
client = OpenAI(api_key=DeepSeek_API_KEY, base_url="https://api.deepseek.com")

# 调用DeepSeek的API，生成回答
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "system", "content": "你是乐于助人的助手，请根据用户的问题给出回答"},
        {"role": "user", "content": "你好，请你介绍一下你自己。"},
    ],
)

# 打印模型最终的响应结果
print(response.choices[0].message.content)

Plaintext
你好！我是一个乐于助人的AI助手，随时准备为你提供帮助。我可以回答各种问题、提供建议、协助解决问题，或者陪你聊天。无论是学习、工作还是日常生活，我都会尽力为你提供有用的信息和支持。

我的知识涵盖了多个领域，包括但不限于科技、历史、文化、健康、编程等。如果你有任何疑问或需要帮助，随时告诉我！ 😊

你想了解关于我的具体方面，还是需要其他帮助呢？

接下来我们要考虑的是，对于这样一个DeepSeek官方的API，如何接入到LangGraph中呢？其实非常简单，我们只需要使用LangChain中的一个DeepSeek组件即可向像述代码一样，直接使用相同的DeepSeek API KEY与大模型进行交互。因此，我们首先需要安装LangChain的DeepSeek组件，安装命令如下：

Python
! pip install langchain-deepseek

Plaintext
Collecting langchain-deepseek
Downloading langchain_deepseek-0.1.3-py3-none-any.whl.metadata (1.1 kB)
Requirement already satisfied: langchain-core<1.0.0,>=0.3.47 in /root/anaconda3/lib/python3.12/site-packages (from langchain-deepseek) (0.3.64)
Collecting langchain-openai<1.0.0,>=0.3.9 (from langchain-deepseek)
Downloading langchain_openai-0.3.21-py3-none-any.whl.metadata (2.3 kB)
Requirement already satisfied: langsmith<0.4,>=0.3.45 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (0.3.45)
Requirement already satisfied: tenacity!=8.4.0,<10.0.0,>=8.1.0 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (8.5.0)
Requirement already satisfied: jsonpatch<2.0,>=1.33 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (1.33)
Requirement already satisfied: PyYAML>=5.3 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (6.0.1)
Requirement already satisfied: packaging<25,>=23.2 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (24.1)
Requirement already satisfied: typing-extensions>=4.7 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (4.13.2)
Requirement already satisfied: pydantic>=2.7.4 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (2.11.4)
Requirement already satisfied: openai<2.0.0,>=1.68.2 in /root/anaconda3/lib/python3.12/site-packages (from langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (1.78.1)
Requirement already satisfied: tiktoken<1,>=0.7 in /root/anaconda3/lib/python3.12/site-packages (from langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (0.8.0)
Requirement already satisfied: jsonpointer>=1.9 in /root/anaconda3/lib/python3.12/site-packages (from jsonpatch<2.0,>=1.33->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (2.1)
Requirement already satisfied: httpx<1,>=0.23.0 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (0.28.1)
Requirement already satisfied: orjson<4.0.0,>=3.9.14 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (3.10.13)
Requirement already satisfied: requests<3,>=2 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (2.32.3)
Requirement already satisfied: requests-toolbelt<2.0.0,>=1.0.0 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (1.0.0)
Requirement already satisfied: zstandard<0.24.0,>=0.23.0 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (0.23.0)
Requirement already satisfied: anyio<5,>=3.5.0 in /root/anaconda3/lib/python3.12/site-packages (from openai<2.0.0,>=1.68.2->langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (4.8.0)
Requirement already satisfied: distro<2,>=1.7.0 in /root/anaconda3/lib/python3.12/site-packages (from openai<2.0.0,>=1.68.2->langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (1.9.0)
Requirement already satisfied: jiter<1,>=0.4.0 in /root/anaconda3/lib/python3.12/site-packages (from openai<2.0.0,>=1.68.2->langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (0.8.2)
Requirement already satisfied: sniffio in /root/anaconda3/lib/python3.12/site-packages (from openai<2.0.0,>=1.68.2->langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (1.3.0)
Requirement already satisfied: tqdm>4 in /root/anaconda3/lib/python3.12/site-packages (from openai<2.0.0,>=1.68.2->langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (4.66.4)
Requirement already satisfied: annotated-types>=0.6.0 in /root/anaconda3/lib/python3.12/site-packages (from pydantic>=2.7.4->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (0.6.0)
Requirement already satisfied: pydantic-core==2.33.2 in /root/anaconda3/lib/python3.12/site-packages (from pydantic>=2.7.4->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (2.33.2)
Requirement already satisfied: typing-inspection>=0.4.0 in /root/anaconda3/lib/python3.12/site-packages (from pydantic>=2.7.4->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (0.4.0)
Requirement already satisfied: regex>=2022.1.18 in /root/anaconda3/lib/python3.12/site-packages (from tiktoken<1,>=0.7->langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (2023.10.3)
Requirement already satisfied: idna>=2.8 in /root/anaconda3/lib/python3.12/site-packages (from anyio<5,>=3.5.0->openai<2.0.0,>=1.68.2->langchain-openai<1.0.0,>=0.3.9->langchain-deepseek) (3.7)
Requirement already satisfied: certifi in /root/anaconda3/lib/python3.12/site-packages (from httpx<1,>=0.23.0->langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (2024.2.2)
Requirement already satisfied: httpcore==1.* in /root/anaconda3/lib/python3.12/site-packages (from httpx<1,>=0.23.0->langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (1.0.7)
Requirement already satisfied: h11<0.15,>=0.13 in /root/anaconda3/lib/python3.12/site-packages (from httpcore==1.*->httpx<1,>=0.23.0->langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (0.14.0)
Requirement already satisfied: charset-normalizer<4,>=2 in /root/anaconda3/lib/python3.12/site-packages (from requests<3,>=2->langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (2.0.4)
Requirement already satisfied: urllib3<3,>=1.21.1 in /root/anaconda3/lib/python3.12/site-packages (from requests<3,>=2->langsmith<0.4,>=0.3.45->langchain-core<1.0.0,>=0.3.47->langchain-deepseek) (2.2.2)
Downloading langchain_deepseek-0.1.3-py3-none-any.whl (7.1 kB)
Downloading langchain_openai-0.3.21-py3-none-any.whl (65 kB)
?[2K ?[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━?[0m ?[32m65.2/65.2 kB?[0m ?[31m4.9 kB/s?[0m eta ?[36m0:00:00?[0ma ?[36m0:00:09?[0m
?[?25hInstalling collected packages: langchain-openai, langchain-deepseek
Successfully installed langchain-deepseek-0.1.3 langchain-openai-0.3.21
?[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv?[0m?[33m
?[0m

安装好LangChain集成DeepSeek模型的依赖包后，需要通过一个init_chat_model函数来初始化大模型，代码如下：

Python
from langchain.chat_models import init_chat_model

model = init_chat_model(model="deepseek-chat", model_provider="deepseek")

其中model用来指定要使用的模型名称，而model_provider用来指定模型提供者，当写入deepseek时，会自动加载langchain-deepseek的依赖包，并使用在model中指定的模型名称用来进行交互。

Python
question = "你好，请你介绍一下你自己。"

result = model.invoke(question)
print(result.content)

Plaintext
你好！我是 **DeepSeek Chat**，由深度求索公司（DeepSeek）研发的一款智能AI助手。我可以帮助你解答各种问题，包括学习、工作、编程、写作、生活百科等多个领域。

### **我的特点：**
✅ **免费使用**：目前无需付费，你可以随时向我提问！
✅ **知识丰富**：我的知识截止到 **2024年7月**，可以为你提供较新的信息。
✅ **超长上下文支持**：可以处理 **128K** 长度的文本，适合分析长文档或复杂问题。
✅ **文件阅读**：支持上传 **PDF、Word、Excel、PPT、TXT** 等文件，并从中提取信息进行分析。
✅ **多语言能力**：可以用中文、英文等多种语言交流，帮助你翻译或学习外语。
✅ **编程助手**：能写代码、调试、优化算法，支持Python、C++、Java等多种编程语言。

### **我能帮你做什么？**
📖 **学习辅导**：解题思路、论文写作、知识点讲解
💼 **工作效率**：写邮件、做PPT、总结报告
💡 **创意灵感**：写故事、起名字、头脑风暴
📊 **数据分析**：处理表格、绘制图表、解读数据
🔧 **技术支持**：代码调试、算法优化、技术咨询

你可以随时向我提问，我会尽力提供最准确、有用的回答！😊 有什么我可以帮你的吗？

Python
result

Plaintext
AIMessage(content='你好！我是 **DeepSeek Chat**，由深度求索公司（DeepSeek）研发的一款智能AI助手。我可以帮助你解答各种问题，包括学习、工作、编程、写作、生活百科等多个领域。 \n\n### **我的特点：** \n✅ **免费使用**：目前无需付费，你可以随时向我提问！ \n✅ **知识丰富**：我的知识截止到 **2024年7月**，可以为你提供较新的信息。 \n✅ **超长上下文支持**：可以处理 **128K** 长度的文本，适合分析长文档或复杂问题。 \n✅ **文件阅读**：支持上传 **PDF、Word、Excel、PPT、TXT** 等文件，并从中提取信息进行分析。 \n✅ **多语言能力**：可以用中文、英文等多种语言交流，帮助你翻译或学习外语。 \n✅ **编程助手**：能写代码、调试、优化算法，支持Python、C++、Java等多种编程语言。 \n\n### **我能帮你做什么？** \n📖 **学习辅导**：解题思路、论文写作、知识点讲解 \n💼 **工作效率**：写邮件、做PPT、总结报告 \n💡 **创意灵感**：写故事、起名字、头脑风暴 \n📊 **数据分析**：处理表格、绘制图表、解读数据 \n🔧 **技术支持**：代码调试、算法优化、技术咨询 \n\n你可以随时向我提问，我会尽力提供最准确、有用的回答！😊 有什么我可以帮你的吗？', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 311, 'prompt_tokens': 9, 'total_tokens': 320, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 0}, 'prompt_cache_hit_tokens': 0, 'prompt_cache_miss_tokens': 9}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0425fp8', 'id': 'e3323dc0-46fa-497b-aaf5-83596184f8b2', 'service_tier': None, 'finish_reason': 'stop', 'logprobs': None}, id='run--9f652093-6feb-484f-99f7-0c30e33290ab-0', usage_metadata={'input_tokens': 9, 'output_tokens': 311, 'total_tokens': 320, 'input_token_details': {'cache_read': 0}, 'output_token_details': {}})

Python
model = init_chat_model(model="deepseek-reasoner", model_provider="deepseek")

result = model.invoke(question)
print(result.content)

Plaintext
你好呀！👋我是 **DeepSeek-R1**，由中国的人工智能公司「深度求索」研发。你可以把我当作一个聪明、热心、24小时在线的文字助手～✨

---

### 🌟 我的特点：
- **知识丰富**：截至2024年7月前的各种知识我都有所了解，无论是学习、工作、生活，还是科技、历史、娱乐，我都能帮上忙。
- **超长记忆力**：可以处理长达128K上下文的内容，也就是说，你发给我超长的文件或对话，我也能理解清楚！
- **文件助手**：支持你上传 **PDF、Word、PPT、Excel、图片** 等文件，我能帮你阅读、总结、提取重点。
- **免费使用**（目前是哦！🎉）而且**无需登录**也能畅快聊天（不过登录后能保存历史记录更方便～）

---

### 🛠 我能做什么？
✅ 解答问题（学习、考试、百科）
✅ 写作助手（作文、小说、文案、邮件）
✅ 编程帮手（写代码、查bug、讲算法）
✅ 文件处理（读论文、总结报告、提取数据）
✅ 生活助手（旅游攻略、美食推荐、情感建议）

---

### 🧠 我的“小限制”：
- 目前是**纯文字模型**，不支持语音或识图（但你可以上传图片文件，我能读取里面的文字）。
- 我的知识截止在 **2024年7月**，之后的新事件就不太清楚啦。
- 我不是真人，但我会尽力用温暖、贴心的方式和你交流 ❤️

---

不管你是学生、上班族、创作者，还是只是好奇想聊聊天——我都在这儿等你！😊
现在，有什么我可以帮你的吗？比如学习上的难题、工作中的任务，或者生活里的小烦恼？

Python
result.additional_kwargs

Plaintext
{'refusal': None,
'reasoning_content': '嗯，用户发来一句简单的问候和自我介绍请求。看起来像是初次接触的破冰场景，可能刚打开聊天界面或者第一次使用这类AI助手。\n\n用户语气礼貌但比较笼统，没有具体需求指向。这种开放式问题通常有两种可能：一是真的想了解AI的功能边界，二是测试性提问看AI如何反应。考虑到中文互联网环境里常有用户用“介绍自己”测试机器人，需要同时做好信息传递和破冰互动。\n\n回复策略上应该突出三点：明确身份属性（非人类但能做什么）、消除陌生感（用表情符号和分段提升可读性）、引导后续对话（结尾提问）。要避免机械的术语堆砌，比如不说“基于Transformer架构”而说“能帮你查资料”。\n\n用户没透露任何个人信息，就用中性称呼。最后那个🎉表情可能有点过，不过新用户首次互动需要点活泼感——要是商务场景用户应该会直接问专业问题吧。结尾提问选“学习/工作/生活”三个维度覆盖常见场景，比单问“有什么可以帮你”更易触发具体需求。'}

Python
result.additional_kwargs['reasoning_content']

Plaintext
'嗯，用户发来一句简单的问候和自我介绍请求。看起来像是初次接触的破冰场景，可能刚打开聊天界面或者第一次使用这类AI助手。\n\n用户语气礼貌但比较笼统，没有具体需求指向。这种开放式问题通常有两种可能：一是真的想了解AI的功能边界，二是测试性提问看AI如何反应。考虑到中文互联网环境里常有用户用“介绍自己”测试机器人，需要同时做好信息传递和破冰互动。\n\n回复策略上应该突出三点：明确身份属性（非人类但能做什么）、消除陌生感（用表情符号和分段提升可读性）、引导后续对话（结尾提问）。要避免机械的术语堆砌，比如不说“基于Transformer架构”而说“能帮你查资料”。\n\n用户没透露任何个人信息，就用中性称呼。最后那个🎉表情可能有点过，不过新用户首次互动需要点活泼感——要是商务场景用户应该会直接问专业问题吧。结尾提问选“学习/工作/生活”三个维度覆盖常见场景，比单问“有什么可以帮你”更易触发具体需求。'

这里可以看到，仅仅通过两行代码，我们便可以在LangChain中顺利调用DeepSeek模型，并得到模型的响应结果。相较于使用DeepSeek的API，使用LangChain调用模型无疑是更加简单的。同时，不仅仅是DeepSeek模型，LangChain还支持其他很多大模型，如OpenAI、Qwen、Gemini等，我们只需要在init_chat_model函数中指定不同的模型名称，就可以调用不同的模型。其工作的原理是这样的：

理解了这个基本原理，如果大家想在用LangGraph进行开发时使用其他大模型如Qwen3系列，则只需要先获取到Qwen3模型的API_KEY，然后安装Tongyi Qwen的第三方依赖包，即可同样通过init_chat_model函数来初始化模型，并调用invoke方法来得到模型的响应结果。关于LangChain都支持哪些大模型以及每个模型对应的是哪个第三方依赖包，大家可以在LangChain的官方文档中找到，访问链接为：https://python.langchain.com/docs/integrations/chat/

2. 借助预构建的智能体API快速创建LangGraph智能体

正如开篇所说，我们先尝试用最高层的API快速构建智能体，然后在后续的课程讲解中，将逐渐深入到其背后的底层原理及更底层的API进行讲解。

在LangGraph所有的顶层API中，最常见的预构建智能体API就是create_react_agent，借助该API，开发者仅需将大模型、工具和提示词模板（可选）以参数形式带入，即可快速创建一个拥有完整功能、并且能够串联并联调用多个外部工具的智能体。

创建自定义工具

Python
import requests,json
import os
from dotenv import load_dotenv
load_dotenv(override=True)

Plaintext
True

Python
def get_weather(loc):
    """
    查询即时天气函数
    :param loc: 必要参数，字符串类型，用于表示查询天气的具体城市名称，\
    注意，中国的城市需要用对应城市的英文名称代替，例如如果需要查询北京市天气，则loc参数需要输入'Beijing'；
    :return：OpenWeather API查询即时天气的结果，具体URL请求地址为：https://api.openweathermap.org/data/2.5/weather\
    返回结果对象类型为解析之后的JSON格式对象，并用字符串形式进行表示，其中包含了全部重要的天气信息
    """
    # Step 1.构建请求
    url = "https://api.openweathermap.org/data/2.5/weather"

    # Step 2.设置查询参数
    params = {
        "q": loc,
        "appid": os.getenv("OPENWEATHER_API_KEY"),    # 输入API key
        "units": "metric",            # 使用摄氏度而不是华氏度
        "lang":"zh_cn"                # 输出语言为简体中文
    }

    # Step 3.发送GET请求
    response = requests.get(url, params=params)

    # Step 4.解析响应
    data = response.json()
    return json.dumps(data)

Python
get_weather("Beijing")

Plaintext
'{"coord": {"lon": 116.3972, "lat": 39.9075}, "weather": [{"id": 804, "main": "Clouds", "description": "\\u9634\\uff0c\\u591a\\u4e91", "icon": "04d"}], "base": "stations", "main": {"temp": 30.2, "feels_like": 29.8, "temp_min": 30.2, "temp_max": 30.2, "pressure": 1004, "humidity": 39, "sea_level": 1004, "grnd_level": 999}, "visibility": 10000, "wind": {"speed": 6.78, "deg": 213, "gust": 9.75}, "clouds": {"all": 100}, "dt": 1750927053, "sys": {"country": "CN", "sunrise": 1750884439, "sunset": 1750938415}, "timezone": 28800, "id": 1816670, "name": "Beijing", "cod": 200}'

在LangGraph中，我们可以直接使用上述外部工具带入到LangGraph中创建智能体，不过更为稳妥的形式，是通过一个结构化工具函数来说明外部函数的参数输入（包括参数类型），以确保在实际调用过程中大模型能够准确识别外部函数的参数类型及其实际含义：

Python
from langchain_core.tools import tool
from pydantic import BaseModel, Field

class WeatherQuery(BaseModel):
    loc: str = Field(description="The location name of the city")

@tool(args_schema = WeatherQuery)
def get_weather(loc):
    """
    查询即时天气函数
    :param loc: 必要参数，字符串类型，用于表示查询天气的具体城市名称，\
    注意，中国的城市需要用对应城市的英文名称代替，例如如果需要查询北京市天气，则loc参数需要输入'Beijing'；
    :return：OpenWeather API查询即时天气的结果，具体URL请求地址为：https://api.openweathermap.org/data/2.5/weather\
    返回结果对象类型为解析之后的JSON格式对象，并用字符串形式进行表示，其中包含了全部重要的天气信息
    """
    # Step 1.构建请求
    url = "https://api.openweathermap.org/data/2.5/weather"

    # Step 2.设置查询参数
    params = {
        "q": loc,
        "appid": os.getenv("OPENWEATHER_API_KEY"),    # 输入API key
        "units": "metric",            # 使用摄氏度而不是华氏度
        "lang":"zh_cn"                # 输出语言为简体中文
    }

    # Step 3.发送GET请求
    response = requests.get(url, params=params)

    # Step 4.解析响应
    data = response.json()
    return json.dumps(data)

Python
print(f'''
name: {get_weather.name}
description: {get_weather.description}
arguments: {get_weather.args}
''')

Plaintext
name: get_weather
description: 查询即时天气函数
:param loc: 必要参数，字符串类型，用于表示查询天气的具体城市名称，注意，中国的城市需要用对应城市的英文名称代替，例如如果需要查询北京市天气，则loc参数需要输入'Beijing'；
:return：OpenWeather API查询即时天气的结果，具体URL请求地址为：https://api.openweathermap.org/data/2.5/weather 返回结果对象类型为解析之后的JSON格式对象，并用字符串形式进行表示，其中包含了全部重要的天气信息
arguments: {'loc': {'description': 'The location name of the city', 'title': 'Loc', 'type': 'string'}}

其中Pydantic 提供的 BaseModel 负责定义外部函数的参数结构，相当于是告诉模型外部函数参数名是 loc、类型是字符串（str）、描述为“城市名称”，有了这个结构，模型能更好地理解如何调用该函数。

创建LangGraph智能体

Python
# 封装外部函数列表
tools = [get_weather]

Python
from langchain.chat_models import init_chat_model

model = init_chat_model(model="deepseek-chat", model_provider="deepseek")

Python
from langgraph.prebuilt import create_react_agent

Python
agent = create_react_agent(model=model, tools=tools)

Python
response = agent.invoke({"messages": [{"role": "user", "content": "你好，请介绍下你自己。"}]})

Python
response

Plaintext
{'messages': [HumanMessage(content='你好，请介绍下你自己。', additional_kwargs={}, response_metadata={}, id='2eec6ae4-28f9-4ddf-bf25-e205567edf7e'),
AIMessage(content='你好！我是一个智能助手，可以帮助你解答问题、提供信息、完成任务，或者陪你聊天。无论是学习、工作还是生活中的问题，我都可以尽力协助你。例如：\n\n- **信息查询**：天气、新闻、百科知识等。\n- **任务协助**：日程安排、提醒、计算、翻译等。\n- **创意支持**：写作灵感、代码编写、设计建议等。\n- **娱乐互动**：讲笑话、推荐电影、音乐等。\n\n如果你有任何需求或问题，随时告诉我！ 😊', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 108, 'prompt_tokens': 223, 'total_tokens': 331, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 192}, 'prompt_cache_hit_tokens': 192, 'prompt_cache_miss_tokens': 31}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': 'b3506888-3a23-4202-8eec-12a416642a6e', 'service_tier': None, 'finish_reason': 'stop', 'logprobs': None}, id='run--1c0a066d-1145-4aa5-ab00-041cfb9ee438-0', usage_metadata={'input_tokens': 223, 'output_tokens': 108, 'total_tokens': 331, 'input_token_details': {'cache_read': 192}, 'output_token_details': {}})]}

Python
response["messages"]

Plaintext
[HumanMessage(content='你好，请介绍下你自己。', additional_kwargs={}, response_metadata={}, id='2eec6ae4-28f9-4ddf-bf25-e205567edf7e'),
AIMessage(content='你好！我是一个智能助手，可以帮助你解答问题、提供信息、完成任务，或者陪你聊天。无论是学习、工作还是生活中的问题，我都可以尽力协助你。例如：\n\n- **信息查询**：天气、新闻、百科知识等。\n- **任务协助**：日程安排、提醒、计算、翻译等。\n- **创意支持**：写作灵感、代码编写、设计建议等。\n- **娱乐互动**：讲笑话、推荐电影、音乐等。\n\n如果你有任何需求或问题，随时告诉我！ 😊', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 108, 'prompt_tokens': 223, 'total_tokens': 331, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 192}, 'prompt_cache_hit_tokens': 192, 'prompt_cache_miss_tokens': 31}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': 'b3506888-3a23-4202-8eec-12a416642a6e', 'service_tier': None, 'finish_reason': 'stop', 'logprobs': None}, id='run--1c0a066d-1145-4aa5-ab00-041cfb9ee438-0', usage_metadata={'input_tokens': 223, 'output_tokens': 108, 'total_tokens': 331, 'input_token_details': {'cache_read': 192}, 'output_token_details': {}})]

Python
response["messages"][-1].content

Plaintext
'你好！我是一个智能助手，可以帮助你解答问题、提供信息、完成任务，或者陪你聊天。无论是学习、工作还是生活中的问题，我都可以尽力协助你。例如：\n\n- **信息查询**：天气、新闻、百科知识等。\n- **任务协助**：日程安排、提醒、计算、翻译等。\n- **创意支持**：写作灵感、代码编写、设计建议等。\n- **娱乐互动**：讲笑话、推荐电影、音乐等。\n\n如果你有任何需求或问题，随时告诉我！ 😊'

然后测试外部工具调用

Python
response = agent.invoke({"messages": [{"role": "user", "content": "请问北京今天天气如何？"}]})
print(response["messages"][-1].content)

Plaintext
北京今天的天气情况如下：

- **天气状况**：阴，多云
- **当前温度**：30.22°C
- **体感温度**：30.33°C
- **湿度**：43%
- **气压**：1004 hPa
- **风速**：6.78 m/s，风向为西南方向（213°）
- **能见度**：10公里

请注意防晒和补水，尤其是多云天气下紫外线依然较强。

Python
response['messages']

Plaintext
[HumanMessage(content='请问北京今天天气如何？', additional_kwargs={}, response_metadata={}, id='437024fa-2082-4194-9b58-8a8affb6b9c6'),
AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_0_be105d52-221c-4eb2-bfe1-7e21279fa53f', 'function': {'arguments': '{"loc":"Beijing"}', 'name': 'get_weather'}, 'type': 'function', 'index': 0}], 'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 19, 'prompt_tokens': 222, 'total_tokens': 241, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 192}, 'prompt_cache_hit_tokens': 192, 'prompt_cache_miss_tokens': 30}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': 'e94184ce-4b25-4b5d-9c45-b8cc483ff539', 'service_tier': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run--638f21a3-91dc-4619-8bdd-66ac48d7e611-0', tool_calls=[{'name': 'get_weather', 'args': {'loc': 'Beijing'}, 'id': 'call_0_be105d52-221c-4eb2-bfe1-7e21279fa53f', 'type': 'tool_call'}], usage_metadata={'input_tokens': 222, 'output_tokens': 19, 'total_tokens': 241, 'input_token_details': {'cache_read': 192}, 'output_token_details': {}}),
ToolMessage(content='{"coord": {"lon": 116.3972, "lat": 39.9075}, "weather": [{"id": 804, "main": "Clouds", "description": "\\u9634\\uff0c\\u591a\\u4e91", "icon": "04d"}], "base": "stations", "main": {"temp": 30.22, "feels_like": 30.33, "temp_min": 30.22, "temp_max": 30.22, "pressure": 1004, "humidity": 43, "sea_level": 1004, "grnd_level": 999}, "visibility": 10000, "wind": {"speed": 6.78, "deg": 213, "gust": 9.75}, "clouds": {"all": 100}, "dt": 1750930167, "sys": {"country": "CN", "sunrise": 1750884439, "sunset": 1750938415}, "timezone": 28800, "id": 1816670, "name": "Beijing", "cod": 200}', name='get_weather', id='f9b7d2d6-05da-496f-b075-22cbd8f9ee19', tool_call_id='call_0_be105d52-221c-4eb2-bfe1-7e21279fa53f'),
AIMessage(content='北京今天的天气情况如下：\n\n- **天气状况**：阴，多云\n- **当前温度**：30.22°C\n- **体感温度**：30.33°C\n- **湿度**：43%\n- **气压**：1004 hPa\n- **风速**：6.78 m/s，风向为西南方向（213°）\n- **能见度**：10公里\n\n请注意防晒和补水，尤其是多云天气下紫外线依然较强。', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 99, 'prompt_tokens': 496, 'total_tokens': 595, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 192}, 'prompt_cache_hit_tokens': 192, 'prompt_cache_miss_tokens': 304}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': '00c08d2e-f7aa-4980-b669-cef67c497dca', 'service_tier': None, 'finish_reason': 'stop', 'logprobs': None}, id='run--fd106677-6489-4a64-8935-210072b57014-0', usage_metadata={'input_tokens': 496, 'output_tokens': 99, 'total_tokens': 595, 'input_token_details': {'cache_read': 192}, 'output_token_details': {}})]

而这背后其实就是一次完整的Function calling调用流程：

Python
len(response['messages'])

Plaintext
4

Python
response['messages'][0]

Plaintext
HumanMessage(content='请问北京今天天气如何？', additional_kwargs={}, response_metadata={}, id='437024fa-2082-4194-9b58-8a8affb6b9c6')

Python
response['messages'][1]

Plaintext
AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_0_be105d52-221c-4eb2-bfe1-7e21279fa53f', 'function': {'arguments': '{"loc":"Beijing"}', 'name': 'get_weather'}, 'type': 'function', 'index': 0}], 'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 19, 'prompt_tokens': 222, 'total_tokens': 241, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 192}, 'prompt_cache_hit_tokens': 192, 'prompt_cache_miss_tokens': 30}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': 'e94184ce-4b25-4b5d-9c45-b8cc483ff539', 'service_tier': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run--638f21a3-91dc-4619-8bdd-66ac48d7e611-0', tool_calls=[{'name': 'get_weather', 'args': {'loc': 'Beijing'}, 'id': 'call_0_be105d52-221c-4eb2-bfe1-7e21279fa53f', 'type': 'tool_call'}], usage_metadata={'input_tokens': 222, 'output_tokens': 19, 'total_tokens': 241, 'input_token_details': {'cache_read': 192}, 'output_token_details': {}})

Python
response['messages'][2]

Plaintext
ToolMessage(content='{"coord": {"lon": 116.3972, "lat": 39.9075}, "weather": [{"id": 804, "main": "Clouds", "description": "\\u9634\\uff0c\\u591a\\u4e91", "icon": "04d"}], "base": "stations", "main": {"temp": 30.22, "feels_like": 30.33, "temp_min": 30.22, "temp_max": 30.22, "pressure": 1004, "humidity": 43, "sea_level": 1004, "grnd_level": 999}, "visibility": 10000, "wind": {"speed": 6.78, "deg": 213, "gust": 9.75}, "clouds": {"all": 100}, "dt": 1750930167, "sys": {"country": "CN", "sunrise": 1750884439, "sunset": 1750938415}, "timezone": 28800, "id": 1816670, "name": "Beijing", "cod": 200}', name='get_weather', id='f9b7d2d6-05da-496f-b075-22cbd8f9ee19', tool_call_id='call_0_be105d52-221c-4eb2-bfe1-7e21279fa53f')

Python
response['messages'][3]

Plaintext
AIMessage(content='北京今天的天气情况如下：\n\n- **天气状况**：阴，多云\n- **当前温度**：30.22°C\n- **体感温度**：30.33°C\n- **湿度**：43%\n- **气压**：1004 hPa\n- **风速**：6.78 m/s，风向为西南方向（213°）\n- **能见度**：10公里\n\n请注意防晒和补水，尤其是多云天气下紫外线依然较强。', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 99, 'prompt_tokens': 496, 'total_tokens': 595, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 192}, 'prompt_cache_hit_tokens': 192, 'prompt_cache_miss_tokens': 304}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': '00c08d2e-f7aa-4980-b669-cef67c497dca', 'service_tier': None, 'finish_reason': 'stop', 'logprobs': None}, id='run--fd106677-6489-4a64-8935-210072b57014-0', usage_metadata={'input_tokens': 496, 'output_tokens': 99, 'total_tokens': 595, 'input_token_details': {'cache_read': 192}, 'output_token_details': {}})

更加完整的Function calling执行链路

3. LangGraph React Agent外部工具响应形式

借助LangGraph React Agent，无需任何额外设置，即可完成多工具串联和并联调用。

Python
response = agent.invoke({"messages": [{"role": "user", "content": "请问北京和杭州今天哪里更热？"}]})

Python
response

Plaintext
{'messages': [HumanMessage(content='请问北京和杭州今天哪里更热？', additional_kwargs={}, response_metadata={}, id='61c31341-a015-4991-a828-d26047470913'),
AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_0_428a28c6-a65c-4bcb-b3d2-4ffd7c5b8c2a', 'function': {'arguments': '{"loc": "Beijing"}', 'name': 'get_weather'}, 'type': 'function', 'index': 0}, {'id': 'call_1_8cfb80ca-94b7-4df8-ae01-f1492250f318', 'function': {'arguments': '{"loc": "Hangzhou"}', 'name': 'get_weather'}, 'type': 'function', 'index': 1}], 'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 41, 'prompt_tokens': 344, 'total_tokens': 385, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 320}, 'prompt_cache_hit_tokens': 320, 'prompt_cache_miss_tokens': 24}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': 'f696ffdf-f12d-444d-9ac4-a5869eb92bb5', 'service_tier': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run--0d5b607b-ecea-4076-a1d6-882b4c61875b-0', tool_calls=[{'name': 'get_weather', 'args': {'loc': 'Beijing'}, 'id': 'call_0_428a28c6-a65c-4bcb-b3d2-4ffd7c5b8c2a', 'type': 'tool_call'}, {'name': 'get_weather', 'args': {'loc': 'Hangzhou'}, 'id': 'call_1_8cfb80ca-94b7-4df8-ae01-f1492250f318', 'type': 'tool_call'}], usage_metadata={'input_tokens': 344, 'output_tokens': 41, 'total_tokens': 385, 'input_token_details': {'cache_read': 320}, 'output_token_details': {}}),
ToolMessage(content='{"coord": {"lon": 116.3972, "lat": 39.9075}, "weather": [{"id": 804, "main": "Clouds", "description": "\\u9634\\uff0c\\u591a\\u4e91", "icon": "04d"}], "base": "stations", "main": {"temp": 28.42, "feels_like": 28.04, "temp_min": 28.42, "temp_max": 28.42, "pressure": 1004, "humidity": 40, "sea_level": 1004, "grnd_level": 999}, "visibility": 10000, "wind": {"speed": 7.24, "deg": 214, "gust": 10.16}, "clouds": {"all": 97}, "dt": 1750930823, "sys": {"country": "CN", "sunrise": 1750884439, "sunset": 1750938415}, "timezone": 28800, "id": 1816670, "name": "Beijing", "cod": 200}', name='get_weather', id='b7b3590b-b843-45e3-85bb-cb5f5a2486eb', tool_call_id='call_0_428a28c6-a65c-4bcb-b3d2-4ffd7c5b8c2a'),
ToolMessage(content='{"coord": {"lon": 120.1614, "lat": 30.2937}, "weather": [{"id": 501, "main": "Rain", "description": "\\u4e2d\\u96e8", "icon": "10d"}], "base": "stations", "main": {"temp": 31.9, "feels_like": 38.9, "temp_min": 31.9, "temp_max": 31.9, "pressure": 1004, "humidity": 75, "sea_level": 1004, "grnd_level": 1002}, "visibility": 10000, "wind": {"speed": 1.18, "deg": 26, "gust": 3.61}, "rain": {"1h": 1.5}, "clouds": {"all": 95}, "dt": 1750931166, "sys": {"country": "CN", "sunrise": 1750885155, "sunset": 1750935893}, "timezone": 28800, "id": 1808926, "name": "Hangzhou", "cod": 200}', name='get_weather', id='2fe0745d-87e6-4ac7-ac22-29b0411dfd9a', tool_call_id='call_1_8cfb80ca-94b7-4df8-ae01-f1492250f318'),
AIMessage(content='根据最新的天气数据：\n\n- **北京**：当前温度为 **28.42°C**，天气多云，体感温度为 **28.04°C**，湿度为 **40%**。\n- **杭州**：当前温度为 **31.9°C**，天气中雨，体感温度为 **38.9°C**，湿度为 **75%**。\n\n**结论**：杭州比北京更热，尤其是体感温度更高，且湿度较大。', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 98, 'prompt_tokens': 900, 'total_tokens': 998, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 384}, 'prompt_cache_hit_tokens': 384, 'prompt_cache_miss_tokens': 516}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': '1c59465c-51fe-47d3-9974-8249cbdb1508', 'service_tier': None, 'finish_reason': 'stop', 'logprobs': None}, id='run--a0079b51-a5c2-48b9-bb08-b36ab4c2cc6f-0', usage_metadata={'input_tokens': 900, 'output_tokens': 98, 'total_tokens': 998, 'input_token_details': {'cache_read': 384}, 'output_token_details': {}})]}

Python
from langchain_core.tools import tool
from pydantic import BaseModel, Field

class Write_Query(BaseModel):
    content: str = Field(description="需要写入文档的具体内容")

@tool(args_schema = Write_Query)
def write_file(content: str) -> str:
    """
    将指定内容写入本地文件。
    :param content: 必要参数，字符串类型，用于表示需要写入文档的具体内容。
    :return：是否成功写入
    """
    return "已成功写入本地文件。"

Python
# 封装外部函数列表
tools = [get_weather, write_file]

Python
agent = create_react_agent(model=model, tools=tools)

Python
response = agent.invoke({"messages": [{"role": "user", "content": "你好，请帮我查询北京和杭州的天气，并将其写入本地文件中。"}]})

Python
print(response["messages"][-1].content)

Plaintext
我已经将北京和杭州的天气信息写入本地文件中。以下是内容摘要：

**北京天气：**
- 天气状况：阴，多云
- 温度：28.42°C
- 体感温度：28.04°C
- 湿度：40%
- 风速：7.24 m/s

**杭州天气：**
- 天气状况：中雨
- 温度：31.9°C
- 体感温度：38.9°C
- 湿度：75%
- 风速：1.18 m/s

文件已成功保存。

Python
len(response['messages'])

Plaintext
7

Python
response['messages']

Plaintext
[HumanMessage(content='你好，请帮我查询北京和杭州的天气，并将其写入本地文件中。', additional_kwargs={}, response_metadata={}, id='7e5663b1-7c16-4c5d-a71a-c04bea24a845'),
AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_0_5d676409-f24e-45b6-a1c6-feab6b607025', 'function': {'arguments': '{"loc": "Beijing"}', 'name': 'get_weather'}, 'type': 'function', 'index': 0}, {'id': 'call_1_e0bf4f8c-80ae-4589-b289-4b5a1961795a', 'function': {'arguments': '{"loc": "Hangzhou"}', 'name': 'get_weather'}, 'type': 'function', 'index': 1}], 'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 42, 'prompt_tokens': 351, 'total_tokens': 393, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 256}, 'prompt_cache_hit_tokens': 256, 'prompt_cache_miss_tokens': 95}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': 'aa66d5f6-9822-4fa8-9893-54e1069bb188', 'service_tier': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run--e2f601f1-2b61-44cb-ae0a-53032e25b34b-0', tool_calls=[{'name': 'get_weather', 'args': {'loc': 'Beijing'}, 'id': 'call_0_5d676409-f24e-45b6-a1c6-feab6b607025', 'type': 'tool_call'}, {'name': 'get_weather', 'args': {'loc': 'Hangzhou'}, 'id': 'call_1_e0bf4f8c-80ae-4589-b289-4b5a1961795a', 'type': 'tool_call'}], usage_metadata={'input_tokens': 351, 'output_tokens': 42, 'total_tokens': 393, 'input_token_details': {'cache_read': 256}, 'output_token_details': {}}),
ToolMessage(content='{"coord": {"lon": 116.3972, "lat": 39.9075}, "weather": [{"id": 804, "main": "Clouds", "description": "\\u9634\\uff0c\\u591a\\u4e91", "icon": "04d"}], "base": "stations", "main": {"temp": 28.42, "feels_like": 28.04, "temp_min": 28.42, "temp_max": 28.42, "pressure": 1004, "humidity": 40, "sea_level": 1004, "grnd_level": 999}, "visibility": 10000, "wind": {"speed": 7.24, "deg": 214, "gust": 10.16}, "clouds": {"all": 97}, "dt": 1750930952, "sys": {"country": "CN", "sunrise": 1750884439, "sunset": 1750938415}, "timezone": 28800, "id": 1816670, "name": "Beijing", "cod": 200}', name='get_weather', id='4be5d0a3-c9bf-44e8-a2fa-b7d6ad0665c7', tool_call_id='call_0_5d676409-f24e-45b6-a1c6-feab6b607025'),
ToolMessage(content='{"coord": {"lon": 120.1614, "lat": 30.2937}, "weather": [{"id": 501, "main": "Rain", "description": "\\u4e2d\\u96e8", "icon": "10d"}], "base": "stations", "main": {"temp": 31.9, "feels_like": 38.9, "temp_min": 31.9, "temp_max": 31.9, "pressure": 1004, "humidity": 75, "sea_level": 1004, "grnd_level": 1002}, "visibility": 10000, "wind": {"speed": 1.18, "deg": 26, "gust": 3.61}, "rain": {"1h": 1.54}, "clouds": {"all": 95}, "dt": 1750930851, "sys": {"country": "CN", "sunrise": 1750885155, "sunset": 1750935893}, "timezone": 28800, "id": 1808926, "name": "Hangzhou", "cod": 200}', name='get_weather', id='b9a422ed-b600-4c9e-8623-ad94eecf9742', tool_call_id='call_1_e0bf4f8c-80ae-4589-b289-4b5a1961795a'),
AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_0_1c081c7f-2cf1-4c31-a7f7-01b9702b2a8c', 'function': {'arguments': '{"content":"北京天气：\\n- 天气状况：阴，多云\\n- 温度：28.42°C\\n- 体感温度：28.04°C\\n- 湿度：40%\\n- 风速：7.24 m/s\\n\\n杭州天气：\\n- 天气状况：中雨\\n- 温度：31.9°C\\n- 体感温度：38.9°C\\n- 湿度：75%\\n- 风速：1.18 m/s"}', 'name': 'write_file'}, 'type': 'function', 'index': 0}], 'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 122, 'prompt_tokens': 907, 'total_tokens': 1029, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 320}, 'prompt_cache_hit_tokens': 320, 'prompt_cache_miss_tokens': 587}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': '7197435e-d9ba-4e9b-a923-b2686a66f320', 'service_tier': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run--85a07e59-12f2-404f-9c6c-9e3806dd6883-0', tool_calls=[{'name': 'write_file', 'args': {'content': '北京天气：\n- 天气状况：阴，多云\n- 温度：28.42°C\n- 体感温度：28.04°C\n- 湿度：40%\n- 风速：7.24 m/s\n\n杭州天气：\n- 天气状况：中雨\n- 温度：31.9°C\n- 体感温度：38.9°C\n- 湿度：75%\n- 风速：1.18 m/s'}, 'id': 'call_0_1c081c7f-2cf1-4c31-a7f7-01b9702b2a8c', 'type': 'tool_call'}], usage_metadata={'input_tokens': 907, 'output_tokens': 122, 'total_tokens': 1029, 'input_token_details': {'cache_read': 320}, 'output_token_details': {}}),
ToolMessage(content='已成功写入本地文件。', name='write_file', id='6f60980d-07a2-4135-b2db-5077bc3204a0', tool_call_id='call_0_1c081c7f-2cf1-4c31-a7f7-01b9702b2a8c'),
AIMessage(content='我已经将北京和杭州的天气信息写入本地文件中。以下是内容摘要：\n\n**北京天气：**\n- 天气状况：阴，多云\n- 温度：28.42°C\n- 体感温度：28.04°C\n- 湿度：40%\n- 风速：7.24 m/s\n\n**杭州天气：**\n- 天气状况：中雨\n- 温度：31.9°C\n- 体感温度：38.9°C\n- 湿度：75%\n- 风速：1.18 m/s\n\n文件已成功保存。', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 124, 'prompt_tokens': 1219, 'total_tokens': 1343, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 896}, 'prompt_cache_hit_tokens': 896, 'prompt_cache_miss_tokens': 323}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': 'f09baf1b-9ee8-4077-b851-89b511bbffaa', 'service_tier': None, 'finish_reason': 'stop', 'logprobs': None}, id='run--88806651-2000-403d-9afe-c7a44f3d9583-0', usage_metadata={'input_tokens': 1219, 'output_tokens': 124, 'total_tokens': 1343, 'input_token_details': {'cache_read': 896}, 'output_token_details': {}})]

能够看出，此时Agent能够顺利完成多工具的多次调用，并最终完成用户需求。需要注意的是，此时react agent的执行逻辑是先发起了一次并联调用工具的请求，然后第二次再发起调用write函数的调用请求，因此总共是7条消息。

4. LangGraph React Agent接入内置工具流程

当然对于LangGraph智能体来说，除了能够灵活接如自定义工具，还能够接入LangChain丰富的内置工具，快速完成智能体开发。

在 LangChain 框架中，**工具（Tools）**是实现语言模型与外部世界交互的关键机制。LangChain 提供了大量内置与可扩展的工具接口，使得智能体（Agent）能够执行函数调用、访问 API、查询搜索引擎、调用数据库等任务，从而超越纯语言生成的能力，真正实现“能行动的智能体”。LangChain 官方文档将这些工具按照其用途进行了模块化划分，涵盖了以下主要类别：

功能类别	工具名称	简要说明
🔎 搜索工具	TavilySearchResults	快速搜索实时网络信息
	SerpAPIWrapper	基于 SerpAPI 的搜索结果工具
	GoogleSearchAPIWrapper	调用 Google 可编程搜索引擎
🧠 计算工具	PythonREPLTool	执行 Python 表达式并返回结果
	LLMMathTool	结合 LLM 和数学推理能力
	WolframAlphaQueryRun	基于 Wolfram Alpha 的计算引擎
🗂 数据工具	SQLDatabaseToolkit	构建 SQL 数据库查询工具集
	PandasDataframeTool	用于在 Agent 中操作表格数据
🌐 网络/API	RequestsGetTool / RequestsPostTool	执行 HTTP 请求
	BrowserTool / PlaywrightBrowserToolkit	自动化网页浏览与抓取
💾 文件处理	ReadFileTool	读取本地文件内容
	WriteFileTool	写入文本到指定文件中
📚 检索工具	FAISSRetriever	基于向量的文档检索工具
	ChromaRetriever	使用 ChromaDB 的检索器
	ContextualCompressionRetriever	上下文压缩检索器，适合长文档
🧠 LLM 工具	ChatOpenAI / OpenAIFunctionsTool	使用 OpenAI 模型作为工具调用
	ChatAnthropic	Anthropic Claude 模型封装工具
🔧 自定义工具	@tool 装饰器	任意函数可封装为 Agent 可调用工具
	Tool 类继承	自定义更复杂逻辑的工具实现

LangChain工具集：https://python.langchain.com/docs/integrations/tools/

创建带搜索功能的Agent

这里我们尝试借助LangChain内置的Tavily搜索引擎工具，搭建能够进行网络搜索的智能体。首先需要在tavily官网进行注册并获取API-KEY：https://www.tavily.com/

并继续将tavily的API-KEY写到.env文件中：

然后安装对应的库:

Python
!pip install -U langchain-tavily

Plaintext
Requirement already satisfied: langchain-tavily in /root/anaconda3/lib/python3.12/site-packages (0.2.4)
Requirement already satisfied: aiohttp<4.0.0,>=3.11.14 in /root/anaconda3/lib/python3.12/site-packages (from langchain-tavily) (3.11.18)
Requirement already satisfied: langchain<0.4.0,>=0.3.20 in /root/anaconda3/lib/python3.12/site-packages (from langchain-tavily) (0.3.25)
Requirement already satisfied: langchain-core<0.4.0,>=0.3.15 in /root/anaconda3/lib/python3.12/site-packages (from langchain-tavily) (0.3.65)
Requirement already satisfied: mypy<2.0.0,>=1.15.0 in /root/anaconda3/lib/python3.12/site-packages (from langchain-tavily) (1.16.0)
Requirement already satisfied: requests<3.0.0,>=2.32.3 in /root/anaconda3/lib/python3.12/site-packages (from langchain-tavily) (2.32.3)
Requirement already satisfied: aiohappyeyeballs>=2.3.0 in /root/anaconda3/lib/python3.12/site-packages (from aiohttp<4.0.0,>=3.11.14->langchain-tavily) (2.4.4)
Requirement already satisfied: aiosignal>=1.1.2 in /root/anaconda3/lib/python3.12/site-packages (from aiohttp<4.0.0,>=3.11.14->langchain-tavily) (1.2.0)
Requirement already satisfied: attrs>=17.3.0 in /root/anaconda3/lib/python3.12/site-packages (from aiohttp<4.0.0,>=3.11.14->langchain-tavily) (25.3.0)
Requirement already satisfied: frozenlist>=1.1.1 in /root/anaconda3/lib/python3.12/site-packages (from aiohttp<4.0.0,>=3.11.14->langchain-tavily) (1.4.0)
Requirement already satisfied: multidict<7.0,>=4.5 in /root/anaconda3/lib/python3.12/site-packages (from aiohttp<4.0.0,>=3.11.14->langchain-tavily) (6.0.4)
Requirement already satisfied: propcache>=0.2.0 in /root/anaconda3/lib/python3.12/site-packages (from aiohttp<4.0.0,>=3.11.14->langchain-tavily) (0.2.1)
Requirement already satisfied: yarl<2.0,>=1.17.0 in /root/anaconda3/lib/python3.12/site-packages (from aiohttp<4.0.0,>=3.11.14->langchain-tavily) (1.18.3)
Requirement already satisfied: langchain-text-splitters<1.0.0,>=0.3.8 in /root/anaconda3/lib/python3.12/site-packages (from langchain<0.4.0,>=0.3.20->langchain-tavily) (0.3.8)
Requirement already satisfied: langsmith<0.4,>=0.1.17 in /root/anaconda3/lib/python3.12/site-packages (from langchain<0.4.0,>=0.3.20->langchain-tavily) (0.3.45)
Requirement already satisfied: pydantic<3.0.0,>=2.7.4 in /root/anaconda3/lib/python3.12/site-packages (from langchain<0.4.0,>=0.3.20->langchain-tavily) (2.11.4)
Requirement already satisfied: SQLAlchemy<3,>=1.4 in /root/anaconda3/lib/python3.12/site-packages (from langchain<0.4.0,>=0.3.20->langchain-tavily) (2.0.32)
Requirement already satisfied: PyYAML>=5.3 in /root/anaconda3/lib/python3.12/site-packages (from langchain<0.4.0,>=0.3.20->langchain-tavily) (6.0.1)
Requirement already satisfied: tenacity!=8.4.0,<10.0.0,>=8.1.0 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<0.4.0,>=0.3.15->langchain-tavily) (8.5.0)
Requirement already satisfied: jsonpatch<2.0,>=1.33 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<0.4.0,>=0.3.15->langchain-tavily) (1.33)
Requirement already satisfied: packaging<25,>=23.2 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<0.4.0,>=0.3.15->langchain-tavily) (24.1)
Requirement already satisfied: typing-extensions>=4.7 in /root/anaconda3/lib/python3.12/site-packages (from langchain-core<0.4.0,>=0.3.15->langchain-tavily) (4.13.2)
Requirement already satisfied: mypy_extensions>=1.0.0 in /root/anaconda3/lib/python3.12/site-packages (from mypy<2.0.0,>=1.15.0->langchain-tavily) (1.0.0)
Requirement already satisfied: pathspec>=0.9.0 in /root/anaconda3/lib/python3.12/site-packages (from mypy<2.0.0,>=1.15.0->langchain-tavily) (0.10.3)
Requirement already satisfied: charset-normalizer<4,>=2 in /root/anaconda3/lib/python3.12/site-packages (from requests<3.0.0,>=2.32.3->langchain-tavily) (2.0.4)
Requirement already satisfied: idna<4,>=2.5 in /root/anaconda3/lib/python3.12/site-packages (from requests<3.0.0,>=2.32.3->langchain-tavily) (3.7)
Requirement already satisfied: urllib3<3,>=1.21.1 in /root/anaconda3/lib/python3.12/site-packages (from requests<3.0.0,>=2.32.3->langchain-tavily) (1.26.20)
Requirement already satisfied: certifi>=2017.4.17 in /root/anaconda3/lib/python3.12/site-packages (from requests<3.0.0,>=2.32.3->langchain-tavily) (2024.2.2)
Requirement already satisfied: jsonpointer>=1.9 in /root/anaconda3/lib/python3.12/site-packages (from jsonpatch<2.0,>=1.33->langchain-core<0.4.0,>=0.3.15->langchain-tavily) (2.1)
Requirement already satisfied: httpx<1,>=0.23.0 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.1.17->langchain<0.4.0,>=0.3.20->langchain-tavily) (0.28.1)
Requirement already satisfied: orjson<4.0.0,>=3.9.14 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.1.17->langchain<0.4.0,>=0.3.20->langchain-tavily) (3.10.13)
Requirement already satisfied: requests-toolbelt<2.0.0,>=1.0.0 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.1.17->langchain<0.4.0,>=0.3.20->langchain-tavily) (1.0.0)
Requirement already satisfied: zstandard<0.24.0,>=0.23.0 in /root/anaconda3/lib/python3.12/site-packages (from langsmith<0.4,>=0.1.17->langchain<0.4.0,>=0.3.20->langchain-tavily) (0.23.0)
Requirement already satisfied: annotated-types>=0.6.0 in /root/anaconda3/lib/python3.12/site-packages (from pydantic<3.0.0,>=2.7.4->langchain<0.4.0,>=0.3.20->langchain-tavily) (0.6.0)
Requirement already satisfied: pydantic-core==2.33.2 in /root/anaconda3/lib/python3.12/site-packages (from pydantic<3.0.0,>=2.7.4->langchain<0.4.0,>=0.3.20->langchain-tavily) (2.33.2)
Requirement already satisfied: typing-inspection>=0.4.0 in /root/anaconda3/lib/python3.12/site-packages (from pydantic<3.0.0,>=2.7.4->langchain<0.4.0,>=0.3.20->langchain-tavily) (0.4.0)
Requirement already satisfied: greenlet!=0.4.17 in /root/anaconda3/lib/python3.12/site-packages (from SQLAlchemy<3,>=1.4->langchain<0.4.0,>=0.3.20->langchain-tavily) (3.0.1)
Requirement already satisfied: anyio in /root/anaconda3/lib/python3.12/site-packages (from httpx<1,>=0.23.0->langsmith<0.4,>=0.1.17->langchain<0.4.0,>=0.3.20->langchain-tavily) (4.8.0)
Requirement already satisfied: httpcore==1.* in /root/anaconda3/lib/python3.12/site-packages (from httpx<1,>=0.23.0->langsmith<0.4,>=0.1.17->langchain<0.4.0,>=0.3.20->langchain-tavily) (1.0.7)
Requirement already satisfied: h11<0.15,>=0.13 in /root/anaconda3/lib/python3.12/site-packages (from httpcore==1.*->httpx<1,>=0.23.0->langsmith<0.4,>=0.1.17->langchain<0.4.0,>=0.3.20->langchain-tavily) (0.14.0)
Requirement already satisfied: sniffio>=1.1 in /root/anaconda3/lib/python3.12/site-packages (from anyio->httpx<1,>=0.23.0->langsmith<0.4,>=0.1.17->langchain<0.4.0,>=0.3.20->langchain-tavily) (1.3.0)
?[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv?[0m?[33m
?[0m

然后即可导入该工具，并快速将其封装为LangGraph智能体工具：

Python
from langchain_tavily import TavilySearch

Python
search_tool = TavilySearch(max_results=5, topic="general")

Python
tools = [search_tool]

Python
search_agent = create_react_agent(model=model, tools=tools)

Python
response = search_agent.invoke({"messages": [{"role": "user", "content": "请帮我搜索最近OpenAI CEO在访谈中的核心观点。"}]})

Python
print(response["messages"][-1].content)

Plaintext
最近OpenAI CEO Sam Altman在访谈中分享了一些核心观点，以下是主要内容：

1. **追求宏大愿景吸引顶尖人才**：
   - OpenAI的成功始于一个看似疯狂的目标——实现通用人工智能（AGI）。这种独特的使命吸引了世界上最聪明的人才。

2. **创业时机与信念**：
   - Altman认为当前是“技术史上最好的创业时期”，并鼓励创业者保持信念，勇敢面对挑战。
   - 他强调初创企业应在AI巨头的阴影下构建独特的护城河。

3. **未来技术展望**：
   - 他提到未来的人机交互将趋于个性化，界面将“融化”消失，计算机将成为无缝融入生活的智能伙伴。
   - OpenAI的技术路线图包括从GPT-5到机器人时代的发展，期待模型能进行深度推理和实时视频生成。

4. **AI与能源的结合**：
   - Altman对AI和能源（如核聚变）抱有热情，认为智能和能源是通往彻底富足未来的两大技术杠杆。

5. **对创业者的建议**：
   - 他建议创业者关注独特性，避免复制基础模型，利用平台探索尚未满足的市场。

更多详细内容可以参考以下来源：
- [网易新闻访谈](https://www.163.com/dy/article/K2NTSA7805534HHB.html)
- [腾讯新闻访谈](https://news.qq.com/rain/a/20250623A03G6200)

5. LangGraph React Agent限制最大迭代次数

LangChain工具集：https://python.langchain.com/docs/integrations/tools/

Python

5. LangGraph React Agent限制工具调用次数

对于任何全自动的代理，合理控制调用次数都是至关重要的一环，对于LangGraph React Agent来说，我们只需要在Agent运行的时候设置{"recursion_limit": X}，即可限制智能体自主执行任务时的步数。

Python
try:
    response = agent.invoke(
        {"messages": [{"role": "user", "content": "请问北京今天天气如何？"}]},
        {"recursion_limit": 4},
    )
except GraphRecursionError:
    print("Agent stopped due to max iterations.")

Python
response['messages'][-1].content

Plaintext
'北京今天的天气是多云，阴天。当前温度为28.42°C，体感温度为28.04°C。湿度为40%，风速为7.24米/秒，风向为214度。能见度为10000米。'

Python
response['messages']

Plaintext
[HumanMessage(content='请问北京今天天气如何？', additional_kwargs={}, response_metadata={}, id='7bab4f6f-2abd-4caa-a6c7-da79e33562c5'),
AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_0_812fd4ed-a56d-4c59-8c7b-f454a99d0ca3', 'function': {'arguments': '{"loc":"Beijing"}', 'name': 'get_weather'}, 'type': 'function', 'index': 0}], 'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 19, 'prompt_tokens': 341, 'total_tokens': 360, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 320}, 'prompt_cache_hit_tokens': 320, 'prompt_cache_miss_tokens': 21}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': '719434fd-9a71-430c-bc19-0cbf5fc34a00', 'service_tier': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run--f9d3fded-8623-4023-9e23-c71cd01a5e9a-0', tool_calls=[{'name': 'get_weather', 'args': {'loc': 'Beijing'}, 'id': 'call_0_812fd4ed-a56d-4c59-8c7b-f454a99d0ca3', 'type': 'tool_call'}], usage_metadata={'input_tokens': 341, 'output_tokens': 19, 'total_tokens': 360, 'input_token_details': {'cache_read': 320}, 'output_token_details': {}}),
ToolMessage(content='{"coord": {"lon": 116.3972, "lat": 39.9075}, "weather": [{"id": 804, "main": "Clouds", "description": "\\u9634\\uff0c\\u591a\\u4e91", "icon": "04d"}], "base": "stations", "main": {"temp": 28.42, "feels_like": 28.04, "temp_min": 28.42, "temp_max": 28.42, "pressure": 1004, "humidity": 40, "sea_level": 1004, "grnd_level": 999}, "visibility": 10000, "wind": {"speed": 7.24, "deg": 214, "gust": 10.16}, "clouds": {"all": 97}, "dt": 1750931154, "sys": {"country": "CN", "sunrise": 1750884439, "sunset": 1750938415}, "timezone": 28800, "id": 1816670, "name": "Beijing", "cod": 200}', name='get_weather', id='96f4d7d1-2ad2-4f7c-b00f-dfb1523949a2', tool_call_id='call_0_812fd4ed-a56d-4c59-8c7b-f454a99d0ca3'),
AIMessage(content='北京今天的天气是多云，阴天。当前温度为28.42°C，体感温度为28.04°C。湿度为40%，风速为7.24米/秒，风向为214度。能见度为10000米。', additional_kwargs={'refusal': None}, response_metadata={'token_usage': {'completion_tokens': 51, 'prompt_tokens': 615, 'total_tokens': 666, 'completion_tokens_details': None, 'prompt_tokens_details': {'audio_tokens': None, 'cached_tokens': 320}, 'prompt_cache_hit_tokens': 320, 'prompt_cache_miss_tokens': 295}, 'model_name': 'deepseek-chat', 'system_fingerprint': 'fp_8802369eaa_prod0623_fp8_kvcache', 'id': '04f542ba-9aad-48bd-910c-e4f99e257482', 'service_tier': None, 'finish_reason': 'stop', 'logprobs': None}, id='run--75edd7eb-1b92-4889-af0c-4994628f54dc-0', usage_metadata={'input_tokens': 615, 'output_tokens': 51, 'total_tokens': 666, 'input_token_details': {'cache_read': 320}, 'output_token_details': {}})]

Python
len(response['messages'])

Plaintext
4

Python
try:
    response = agent.invoke(
        {"messages": [{"role": "user", "content": "请问北京今天天气如何？"}]},
        {"recursion_limit": 2},
    )
except GraphRecursionError:
    print("Agent stopped due to max iterations.")

Python
response['messages']

Plaintext
[HumanMessage(content='请问北京今天天气如何？', additional_kwargs={}, response_metadata={}, id='7e9edf31-84fa-4a42-bad0-2835aa4ca702'),
AIMessage(content='Sorry, need more steps to process this request.', additional_kwargs={}, response_metadata={}, id='run--9ee8035c-977a-492f-b0e1-aba84ae632ab-0')]

Python
response['messages'][-1].content

Plaintext
'Sorry, need more steps to process this request.'

Python