AI大模型应用开发之LangChain构建含Agen流程的RAG架构

AI大模型应用开发里用LangChain构建含Agen流程的RAG架构

随着大模型（LLM）的能力日益强大，检索增强生成（RAG）技术成了提升大模型知识准确性的关键办法。通过检索实时数据和外部文档，模型能够回答更多基于事实的问题，降低“幻觉”出现的概率。

而LangChain的LangGraph能把LLM、RAG、工具调用（Tools）整合到一个智能Agent流程图里，大大提升了问答系统的动态能力。本文通过一个完整实例，展现怎样用LangChain搭建一个“RAG + Agent”的问答系统，代码能直接复用，助力大家快速落地智能应用。

工程架构

llm_env.py          # 初始化LLM
rag_agent.py        # 结合RAG与Agent的主逻辑

初始化LLM

首先通过llm_env.py来初始化一个LLM模型对象，供整个流程使用：

from langchain.chat_models import init_chat_model

llm = init_chat_model("gpt-4o-mini", model_provider="openai")

RAG + Agent系统搭建

导入依赖

import os
import sys
import time

sys.path.append(os.getcwd())

from llm_set import llm_env
from langchain.embeddings import OpenAIEmbeddings
from langchain_postgres import PGVector
from langchain_community.document_loaders import WebBaseLoader
from langchain_core.documents import Document
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langgraph.graph import MessagesState, StateGraph
from langchain_core.tools import tool
from langchain_core.messages import HumanMessage, SystemMessage
from langgraph.prebuilt import ToolNode, tools_condition
from langgraph.graph import END
from langgraph.checkpoint.postgres import PostgresSaver

初始化LLM与Embedding

llm = llm_env.llm

embeddings = OpenAIEmbeddings(model="text-embedding-3-large")

初始化向量数据库

vector_store = PGVector(
    embeddings=embeddings,
    collection_name="my_rag_agent_docs",
    connection="postgresql+psycopg2://postgres:123456@localhost:5433/langchainvector",
)

加载网页文档

url = "https://www.cnblogs.com/chenyishi/p/18926783"
loader = WebBaseLoader(
    web_paths=(url,),
)
docs = loader.load()
for doc in docs:
    doc.metadata["source"] = url

文本分割 & 入库

text_splitter = RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=50)
all_splits = text_splitter.split_documents(docs)

existing = vector_store.similarity_search(url, k=1, filter={"source": url})
if not existing:
    _ = vector_store.add_documents(documents=all_splits)
    print("文档向量化完成")

定义RAG检索工具

通过@tool装饰器，定义一个文档检索工具，供Agent动态调用：

@tool(response_format="content_and_artifact")
def retrieve(query: str) -> tuple[str, dict]:
    """从向量库中检索相关文档。"""
    retrieved_docs = vector_store.similarity_search(query, k=2)
    if not retrieved_docs:
        return "未找到相关文档。", {}
    return "\n\n".join(
        (f"来源: {doc.metadata}\n" f"内容: {doc.page_content}")
        for doc in retrieved_docs
    ), retrieved_docs

定义Agent Graph节点

LLM调用工具节点

def query_or_respond(state: MessagesState):
    llm_with_tools = llm.bind_tools([retrieve])
    response = llm_with_tools.invoke(state["messages"])
    return {"messages": [response]}

工具节点

tools = ToolNode([retrieve])

生成响应节点

def generate(state: MessagesState):
    recent_tool_messages = []
    for message in reversed(state["messages"]):
        if message.type == "tool":
            recent_tool_messages.append(message)
        else:
            break

    tool_messages = recent_tool_messages[::-1]

    system_message_content = "\n\n".join(doc.content for doc in tool_messages)

    conversation_messages = [
        message
        for message in state["messages"]
        if message.type in ("human", "system")
        or (message.type == "ai" and not message.tool_calls)
    ]
    prompt = [SystemMessage(system_message_content)] + conversation_messages

    response = llm.invoke(prompt)
    return {"messages": [response]}

组装Agent流程图

graph_builder = StateGraph(MessagesState)
graph_builder.add_node(query_or_respond)
graph_builder.add_node(tools)
graph_builder.add_node(generate)

graph_builder.set_entry_point("query_or_respond")
graph_builder.add_conditional_edges(
    "query_or_respond",
    tools_condition,
    path_map={END: END, "tools": "tools"},
)
graph_builder.add_edge("tools", "generate")
graph_builder.add_edge("generate", END)

启用Checkpoint & 运行流程

数据库存储器

DB_URI = "postgresql://postgres:123456@localhost:5433/langchaindemo?sslmode=disable"

with PostgresSaver.from_conn_string(DB_URI) as checkpointer:
    checkpointer.setup()

    graph = graph_builder.compile(checkpointer=checkpointer)

启动交互循环

input_thread_id = input("输入thread_id:")
time_str = time.strftime("%Y%m%d", time.localtime())
config = {"configurable": {"thread_id": f"rag-{time_str}-demo-{input_thread_id}"}}

print("输入问题，输入exit退出。")
while True:
    query = input("你: ")
    if query.strip().lower() == "exit":
        break
    response = graph.invoke({"messages": [HumanMessage(content=query)]}, config=config)
    print(response)

总结

本文完整展示了如何运用LangChain + LangGraph，结合：

LLM（大模型）
Embedding检索（RAG）
Agent动态调用工具
流程图编排
Checkpoint存储

构建一个智能问答系统。通过把工具（RAG检索）和Agent机制相结合，能让LLM在需要的时候自主调用检索能力，有效增强对知识的引用能力，解决“幻觉”问题，具有很好的落地应用价值。