31：RAG已死？2026 Agentic RAG 3.0核心升级点全解析

from langchain.chat_models import ChatOpenAI
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter

# 加载长文档
loader = PyPDFLoader("long_document.pdf")
documents = loader.load()

# 文本分割
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=50000,
    chunk_overlap=5000,
    length_function=len
)
chunks = text_splitter.split_documents(documents)

# 使用支持长上下文的模型
model = ChatOpenAI(model="gpt-5.4", max_tokens=1000000)

# 处理长文档
def process_long_document(chunks, query):
    # 构建长上下文提示
    prompt = f"""
    请基于以下文档内容回答问题：
    
    {"\n".join([chunk.page_content for chunk in chunks])}
    
    问题：{query}
    """
    
    # 调用模型
    response = model.predict(prompt)
    return response

# 示例查询
query = "文档中关于Agentic RAG的核心优势是什么？"
result = process_long_document(chunks, query)
print(result)

from langchain.vectorstores import Neo4jVector
from langchain.embeddings import OpenAIEmbeddings
from langchain.document_loaders import UnstructuredImageLoader
from langchain.document_loaders import YouTubeLoader

# 初始化嵌入模型
embeddings = OpenAIEmbeddings(model="text-embedding-3-large")

# 加载多模态内容
image_loader = UnstructuredImageLoader("image.jpg")
image_docs = image_loader.load()

youtube_loader = YouTubeLoader.from_youtube_url(
    "https://www.youtube.com/watch?v=example",
    add_video_info=True
)
youtube_docs = youtube_loader.load()

# 构建知识图谱
vectorstore = Neo4jVector.from_documents(
    documents=image_docs + youtube_docs,
    embedding=embeddings,
    url="bolt://localhost:7687",
    username="neo4j",
    password="password"
)

# 多模态检索
def multimodal_retrieval(query):
    # 检索相关内容
    results = vectorstore.similarity_search(
        query=query,
        k=5,
        include_metadata=True
    )
    
    # 整合多模态信息
    context = ""
    for result in results:
        context += f"{result.page_content}\n"
    
    # 生成回答
    prompt = f"""
    基于以下多模态内容回答问题：
    
    {context}
    
    问题：{query}
    """
    
    response = model.predict(prompt)
    return response

# 示例查询
query = "视频中提到的Agentic RAG的关键技术是什么？"
result = multimodal_retrieval(query)
print(result)

from langchain.agents import AgentType, initialize_agent
from langchain.tools import Tool
from langchain.chains import RetrievalQA

# 初始化检索QA链
qa_chain = RetrievalQA.from_chain_type(
    llm=model,
    chain_type="stuff",
    retriever=vectorstore.as_retriever()
)

# 定义检索工具
def retrieve_information(query):
    """检索相关信息"""
    result = qa_chain.run(query)
    return result

# 创建工具列表
tools = [
    Tool(
        name="Retrieval",
        func=retrieve_information,
        description="用于检索与问题相关的信息"
    )
]

# 初始化Agent
agent = initialize_agent(
    tools=tools,
    llm=model,
    agent=AgentType.STRUCTURED_CHAT_ZERO_SHOT_REACT_DESCRIPTION,
    verbose=True
)

# 执行复杂查询
complex_query = "详细解释Agentic RAG 3.0的核心技术，并与传统RAG进行对比"
result = agent.run(complex_query)
print(result)

from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Neo4jVector
from langchain.embeddings import OpenAIEmbeddings
from langchain.chat_models import ChatOpenAI

# 加载法律文档
loader = PyPDFLoader("legal_contract.pdf")
documents = loader.load()

# 文本分割
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=100000,
    chunk_overlap=10000
)
chunks = text_splitter.split_documents(documents)

# 初始化嵌入模型
embeddings = OpenAIEmbeddings(model="text-embedding-3-large")

# 构建知识图谱
vectorstore = Neo4jVector.from_documents(
    documents=chunks,
    embedding=embeddings,
    url="bolt://localhost:7687",
    username="neo4j",
    password="password"
)

# 初始化模型
model = ChatOpenAI(model="gpt-5.4", max_tokens=1000000)

# 法律文档分析函数
def analyze_legal_document(query):
    # 检索相关内容
    results = vectorstore.similarity_search(
        query=query,
        k=3,
        include_metadata=True
    )
    
    # 构建长上下文
    context = ""
    for result in results:
        context += f"{result.page_content}\n"
    
    # 生成回答
    prompt = f"""
    你是一位专业的法律分析师，请基于以下法律文档回答问题：
    
    {context}
    
    问题：{query}
    """
    
    response = model.predict(prompt)
    return response

# 示例查询
query = "合同中关于违约责任的条款有哪些？"
result = analyze_legal_document(query)
print(result)

from langchain.document_loaders import UnstructuredImageLoader
from langchain.document_loaders import PyPDFLoader
from langchain.vectorstores import Neo4jVector
from langchain.embeddings import OpenAIEmbeddings
from langchain.chat_models import ChatOpenAI

# 加载医疗影像
image_loader = UnstructuredImageLoader("medical_image.jpg")
image_docs = image_loader.load()

# 加载医疗报告
report_loader = PyPDFLoader("medical_report.pdf")
report_docs = report_loader.load()

# 初始化嵌入模型
embeddings = OpenAIEmbeddings(model="text-embedding-3-large")

# 构建知识图谱
vectorstore = Neo4jVector.from_documents(
    documents=image_docs + report_docs,
    embedding=embeddings,
    url="bolt://localhost:7687",
    username="neo4j",
    password="password"
)

# 初始化模型
model = ChatOpenAI(model="gpt-5.4", max_tokens=1000000)

# 医疗影像分析函数
def analyze_medical_data(query):
    # 检索相关内容
    results = vectorstore.similarity_search(
        query=query,
        k=3,
        include_metadata=True
    )
    
    # 构建多模态上下文
    context = ""
    for result in results:
        context += f"{result.page_content}\n"
    
    # 生成回答
    prompt = f"""
    你是一位专业的医疗分析师，请基于以下医疗数据回答问题：
    
    {context}
    
    问题：{query}
    """
    
    response = model.predict(prompt)
    return response

# 示例查询
query = "基于影像和报告，患者可能的诊断是什么？"
result = analyze_medical_data(query)
print(result)

from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Neo4jVector
from langchain.embeddings import OpenAIEmbeddings
from langchain.chat_models import ChatOpenAI

# 加载金融报告
loader = PyPDFLoader("financial_report.pdf")
documents = loader.load()

# 文本分割
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=150000,
    chunk_overlap=15000
)
chunks = text_splitter.split_documents(documents)

# 初始化嵌入模型
embeddings = OpenAIEmbeddings(model="text-embedding-3-large")

# 构建知识图谱
vectorstore = Neo4jVector.from_documents(
    documents=chunks,
    embedding=embeddings,
    url="bolt://localhost:7687",
    username="neo4j",
    password="password"
)

# 初始化模型
model = ChatOpenAI(model="gpt-5.4", max_tokens=1000000)

# 金融分析函数
def analyze_financial_data(query):
    # 检索相关内容
    results = vectorstore.similarity_search(
        query=query,
        k=3,
        include_metadata=True
    )
    
    # 构建长上下文
    context = ""
    for result in results:
        context += f"{result.page_content}\n"
    
    # 生成回答
    prompt = f"""
    你是一位专业的金融分析师，请基于以下金融数据回答问题：
    
    {context}
    
    问题：{query}
    """
    
    response = model.predict(prompt)
    return response

# 示例查询
query = "基于这份报告，公司未来的投资价值如何？"
result = analyze_financial_data(query)
print(result)

# 安装依赖
pip install langchain openai neo4j python-dotenv

# 配置环境变量
export OPENAI_API_KEY=your-api-key
export NEO4J_URI=bolt://localhost:7687
export NEO4J_USERNAME=neo4j
export NEO4J_PASSWORD=password

# 运行示例
python agentic_rag.py