> ## Documentation Index
> Fetch the complete documentation index at: https://cndoc-langchain.site/llms.txt
> Use this file to discover all available pages before exploring further.
# 使用 LangGraph 构建自定义 RAG 代理
## 概述
在本教程中，我们将使用 LangGraph 构建一个[检索](/oss/python/langchain/retrieval)代理。
LangChain 提供了内置的[代理](/oss/python/langchain/agents)实现，这些实现使用了 [LangGraph](/oss/python/langgraph/overview) 原语。如果需要更深层次的定制，可以直接在 LangGraph 中实现代理。本指南演示了一个检索代理的实现示例。[检索](/oss/python/langchain/retrieval)代理在您希望 LLM 决定是从向量存储中检索上下文还是直接响应用户时非常有用。
在本教程结束时，我们将完成以下工作：
1. 获取并预处理将用于检索的文档。
2. 为这些文档建立语义搜索索引，并为代理创建一个检索器工具。
3. 构建一个代理式 RAG 系统，该系统可以决定何时使用检索器工具。
### 概念
我们将涵盖以下概念：
* 使用[文档加载器](/oss/python/integrations/document_loaders)、[文本分割器](/oss/python/integrations/splitters)、[嵌入](/oss/python/integrations/embeddings)和[向量存储](/oss/python/integrations/vectorstores)进行[检索](/oss/python/langchain/retrieval)
* LangGraph [图 API](/oss/python/langgraph/graph-api)，包括状态、节点、边和条件边。
## 设置
让我们下载所需的包并设置 API 密钥：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
pip install -U langgraph "langchain[openai]" langchain-community langchain-text-splitters bs4
```
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
import getpass
import os
def _set_env(key: str):
if key not in os.environ:
os.environ[key] = getpass.getpass(f"{key}:")
_set_env("OPENAI_API_KEY")
```
注册 LangSmith 以快速发现问题并提升 LangGraph 项目的性能。[LangSmith](https://docs.smith.langchain.com) 允许您使用跟踪数据来调试、测试和监控使用 LangGraph 构建的 LLM 应用。
## 1. 预处理文档
1. 获取将用于我们 RAG 系统的文档。我们将使用 [Lilian Weng 优秀博客](https://lilianweng.github.io/) 中最近的三篇文章。我们将首先使用 `WebBaseLoader` 工具获取页面内容：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from langchain_community.document_loaders import WebBaseLoader
urls = [
"https://lilianweng.github.io/posts/2024-11-28-reward-hacking/",
"https://lilianweng.github.io/posts/2024-07-07-hallucination/",
"https://lilianweng.github.io/posts/2024-04-12-diffusion-video/",
]
docs = [WebBaseLoader(url).load() for url in urls]
```
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
docs[0][0].page_content.strip()[:1000]
```
2. 将获取的文档分割成更小的块，以便索引到我们的向量存储中：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from langchain_text_splitters import RecursiveCharacterTextSplitter
docs_list = [item for sublist in docs for item in sublist]
text_splitter = RecursiveCharacterTextSplitter.from_tiktoken_encoder(
chunk_size=100, chunk_overlap=50
)
doc_splits = text_splitter.split_documents(docs_list)
```
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
doc_splits[0].page_content.strip()
```
## 2. 创建检索器工具
现在我们有了分割后的文档，可以将它们索引到向量存储中，用于语义搜索。
1. 使用内存向量存储和 OpenAI 嵌入：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from langchain_core.vectorstores import InMemoryVectorStore
from langchain_openai import OpenAIEmbeddings
vectorstore = InMemoryVectorStore.from_documents(
documents=doc_splits, embedding=OpenAIEmbeddings()
)
retriever = vectorstore.as_retriever()
```
2. 使用 `@tool` 装饰器创建检索器工具：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from langchain.tools import tool
@tool
def retrieve_blog_posts(query: str) -> str:
"""搜索并返回有关 Lilian Weng 博客文章的信息。"""
docs = retriever.invoke(query)
return "\n\n".join([doc.page_content for doc in docs])
retriever_tool = retrieve_blog_posts
```
3. 测试该工具：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
retriever_tool.invoke({"query": "types of reward hacking"})
```
## 3. 生成查询
现在我们将开始构建代理式 RAG 图的组件（[节点](/oss/python/langgraph/graph-api#nodes)和[边](/oss/python/langgraph/graph-api#edges)）。
请注意，这些组件将操作于 [`MessagesState`](/oss/python/langgraph/graph-api#messagesstate)——一种包含 `messages` 键（其值为[聊天消息](https://python.langchain.com/docs/concepts/messages/)列表）的图状态。
1. 构建一个 `generate_query_or_respond` 节点。它将调用 LLM 根据当前图状态（消息列表）生成响应。根据输入消息，它将决定使用检索器工具进行检索，还是直接响应用户。请注意，我们通过 `.bind_tools` 让聊天模型可以访问我们之前创建的 `retriever_tool`：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from langgraph.graph import MessagesState
from langchain.chat_models import init_chat_model
response_model = init_chat_model("gpt-5.4", temperature=0)
def generate_query_or_respond(state: MessagesState):
"""调用模型根据当前状态生成响应。根据问题，它将决定使用检索器工具进行检索，还是直接响应用户。"""
response = (
response_model
.bind_tools([retriever_tool]).invoke(state["messages"]) # [!code highlight]
)
return {"messages": [response]}
```
2. 在随机输入上尝试：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
input = {"messages": [{"role": "user", "content": "hello!"}]}
generate_query_or_respond(input)["messages"][-1].pretty_print()
```
**输出：**
```
================================== Ai Message ==================================
Hello! How can I help you today?
```
3. 询问一个需要语义搜索的问题：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
input = {
"messages": [
{
"role": "user",
"content": "What does Lilian Weng say about types of reward hacking?",
}
]
}
generate_query_or_respond(input)["messages"][-1].pretty_print()
```
**输出：**
```
================================== Ai Message ==================================
Tool Calls:
retrieve_blog_posts (call_tYQxgfIlnQUDMdtAhdbXNwIM)
Call ID: call_tYQxgfIlnQUDMdtAhdbXNwIM
Args:
query: types of reward hacking
```
## 4. 评估文档
1. 添加一个[条件边](/oss/python/langgraph/graph-api#conditional-edges)——`grade_documents`——以确定检索到的文档是否与问题相关。我们将使用一个具有结构化输出模式 `GradeDocuments` 的模型进行文档评分。`grade_documents` 函数将根据评分决策（`generate_answer` 或 `rewrite_question`）返回要前往的节点名称：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from pydantic import BaseModel, Field
from typing import Literal
GRADE_PROMPT = (
"You are a grader assessing relevance of a retrieved document to a user question. \n "
"Here is the retrieved document: \n\n {context} \n\n"
"Here is the user question: {question} \n"
"If the document contains keyword(s) or semantic meaning related to the user question, grade it as relevant. \n"
"Give a binary score 'yes' or 'no' score to indicate whether the document is relevant to the question."
)
class GradeDocuments(BaseModel): # [!code highlight]
"""使用二元评分对文档进行相关性检查。"""
binary_score: str = Field(
description="相关性评分：如果相关则为 'yes'，如果不相关则为 'no'"
)
grader_model = init_chat_model("gpt-5.4", temperature=0)
def grade_documents(
state: MessagesState,
) -> Literal["generate_answer", "rewrite_question"]:
"""确定检索到的文档是否与问题相关。"""
question = state["messages"][0].content
context = state["messages"][-1].content
prompt = GRADE_PROMPT.format(question=question, context=context)
response = (
grader_model
.with_structured_output(GradeDocuments).invoke( # [!code highlight]
[{"role": "user", "content": prompt}]
)
)
score = response.binary_score
if score == "yes":
return "generate_answer"
else:
return "rewrite_question"
```
2. 在工具响应中包含不相关文档的情况下运行此函数：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from langchain_core.messages import convert_to_messages
input = {
"messages": convert_to_messages(
[
{
"role": "user",
"content": "What does Lilian Weng say about types of reward hacking?",
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "1",
"name": "retrieve_blog_posts",
"args": {"query": "types of reward hacking"},
}
],
},
{"role": "tool", "content": "meow", "tool_call_id": "1"},
]
)
}
grade_documents(input)
```
3. 确认相关文档被正确分类：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
input = {
"messages": convert_to_messages(
[
{
"role": "user",
"content": "What does Lilian Weng say about types of reward hacking?",
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "1",
"name": "retrieve_blog_posts",
"args": {"query": "types of reward hacking"},
}
],
},
{
"role": "tool",
"content": "reward hacking can be categorized into two types: environment or goal misspecification, and reward tampering",
"tool_call_id": "1",
},
]
)
}
grade_documents(input)
```
## 5. 重写问题
1. 构建 `rewrite_question` 节点。检索器工具可能返回潜在的不相关文档，这表明需要改进原始用户问题。为此，我们将调用 `rewrite_question` 节点：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from langchain.messages import HumanMessage
REWRITE_PROMPT = (
"Look at the input and try to reason about the underlying semantic intent / meaning.\n"
"Here is the initial question:"
"\n ------- \n"
"{question}"
"\n ------- \n"
"Formulate an improved question:"
)
def rewrite_question(state: MessagesState):
"""重写原始用户问题。"""
messages = state["messages"]
question = messages[0].content
prompt = REWRITE_PROMPT.format(question=question)
response = response_model.invoke([{"role": "user", "content": prompt}])
return {"messages": [HumanMessage(content=response.content)]}
```
2. 尝试一下：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
input = {
"messages": convert_to_messages(
[
{
"role": "user",
"content": "What does Lilian Weng say about types of reward hacking?",
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "1",
"name": "retrieve_blog_posts",
"args": {"query": "types of reward hacking"},
}
],
},
{"role": "tool", "content": "meow", "tool_call_id": "1"},
]
)
}
response = rewrite_question(input)
print(response["messages"][-1].content)
```
**输出：**
```
What are the different types of reward hacking described by Lilian Weng, and how does she explain them?
```
## 6. 生成答案
1. 构建 `generate_answer` 节点：如果我们通过了评分器检查，就可以根据原始问题和检索到的上下文生成最终答案：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
GENERATE_PROMPT = (
"You are an assistant for question-answering tasks. "
"Use the following pieces of retrieved context to answer the question. "
"If you don't know the answer, just say that you don't know. "
"Use three sentences maximum and keep the answer concise.\n"
"Question: {question} \n"
"Context: {context}"
)
def generate_answer(state: MessagesState):
"""生成答案。"""
question = state["messages"][0].content
context = state["messages"][-1].content
prompt = GENERATE_PROMPT.format(question=question, context=context)
response = response_model.invoke([{"role": "user", "content": prompt}])
return {"messages": [response]}
```
2. 尝试一下：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
input = {
"messages": convert_to_messages(
[
{
"role": "user",
"content": "What does Lilian Weng say about types of reward hacking?",
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "1",
"name": "retrieve_blog_posts",
"args": {"query": "types of reward hacking"},
}
],
},
{
"role": "tool",
"content": "reward hacking can be categorized into two types: environment or goal misspecification, and reward tampering",
"tool_call_id": "1",
},
]
)
}
response = generate_answer(input)
response["messages"][-1].pretty_print()
```
**输出：**
```
================================== Ai Message ==================================
Lilian Weng categorizes reward hacking into two types: environment or goal misspecification, and reward tampering. She considers reward hacking as a broad concept that includes both of these categories. Reward hacking occurs when an agent exploits flaws or ambiguities in the reward function to achieve high rewards without performing the intended behaviors.
```
## 7. 组装图
现在我们将所有节点和边组装成一个完整的图：
* 从 `generate_query_or_respond` 开始，确定是否需要调用 `retriever_tool`
* 使用 `tools_condition` 路由到下一步：
* 如果 `generate_query_or_respond` 返回了 `tool_calls`，则调用 `retriever_tool` 检索上下文
* 否则，直接响应用户
* 评估检索到的文档内容与问题的相关性（`grade_documents`）并路由到下一步：
* 如果不相关，使用 `rewrite_question` 重写问题，然后再次调用 `generate_query_or_respond`
* 如果相关，则进入 `generate_answer`，并使用包含检索到的文档上下文的 [`ToolMessage`](https://reference.langchain.com/python/langchain-core/messages/tool/ToolMessage) 生成最终响应
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from langgraph.graph import StateGraph, START, END
from langgraph.prebuilt import ToolNode, tools_condition
workflow = StateGraph(MessagesState)
# 定义我们将循环使用的节点
workflow.add_node(generate_query_or_respond)
workflow.add_node("retrieve", ToolNode([retriever_tool]))
workflow.add_node(rewrite_question)
workflow.add_node(generate_answer)
workflow.add_edge(START, "generate_query_or_respond")
# 决定是否检索
workflow.add_conditional_edges(
"generate_query_or_respond",
# 评估 LLM 决策（调用 `retriever_tool` 工具还是响应用户）
tools_condition,
{
# 将条件输出转换为图中的节点
"tools": "retrieve",
END: END,
},
)
# 在 `action` 节点被调用后采取的边。
workflow.add_conditional_edges(
"retrieve",
# 评估代理决策
grade_documents,
)
workflow.add_edge("generate_answer", END)
workflow.add_edge("rewrite_question", "generate_query_or_respond")
# 编译
graph = workflow.compile()
```
可视化图：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
from IPython.display import Image, display
display(Image(graph.get_graph().draw_mermaid_png()))
```
## 8. 运行代理式 RAG
现在让我们通过用一个问题运行它来测试完整的图：
```python theme={"theme":{"light":"catppuccin-latte","dark":"catppuccin-mocha"}}
for chunk in graph.stream(
{
"messages": [
{
"role": "user",
"content": "What does Lilian Weng say about types of reward hacking?",
}
]
}
):
for node, update in chunk.items():
print("Update from node", node)
update["messages"][-1].pretty_print()
print("\n\n")
```
**输出：**
```
Update from node generate_query_or_respond
================================== Ai Message ==================================
Tool Calls:
retrieve_blog_posts (call_NYu2vq4km9nNNEFqJwefWKu1)
Call ID: call_NYu2vq4km9nNNEFqJwefWKu1
Args:
query: types of reward hacking
Update from node retrieve
================================= Tool Message ==================================
Name: retrieve_blog_posts
(Note: Some work defines reward tampering as a distinct category of misalignment behavior from reward hacking. But I consider reward hacking as a broader concept here.)
At a high level, reward hacking can be categorized into two types: environment or goal misspecification, and reward tampering.
Why does Reward Hacking Exist?#
Pan et al. (2022) investigated reward hacking as a function of agent capabilities, including (1) model size, (2) action space resolution, (3) observation space noise, and (4) training time. They also proposed a taxonomy of three types of misspecified proxy rewards:
Let's Define Reward Hacking#
Reward shaping in RL is challenging. Reward hacking occurs when an RL agent exploits flaws or ambiguities in the reward function to obtain high rewards without genuinely learning the intended behaviors or completing the task as designed. In recent years, several related concepts have been proposed, all referring to some form of reward hacking:
Update from node generate_answer
================================== Ai Message ==================================
Lilian Weng categorizes reward hacking into two types: environment or goal misspecification, and reward tampering. She considers reward hacking as a broad concept that includes both of these categories. Reward hacking occurs when an agent exploits flaws or ambiguities in the reward function to achieve high rewards without performing the intended behaviors.
```
***
[将这些文档连接](/use-these-docs)到 Claude、VSCode 等，通过 MCP 获取实时答案。
[在 GitHub 上编辑此页面](https://github.com/langchain-ai/docs/edit/main/src/oss/langgraph/agentic-rag.mdx) 或 [提交问题](https://github.com/langchain-ai/docs/issues/new/choose)。