LlamaIndex 教程 2026：构建生产级 RAG 应用

LlamaIndex 是通过 RAG 管道将 LLM 连接到数据的主要框架。

安装

bash
pip install llama-index llama-index-vector-stores-qdrant
pip install llama-index-embeddings-openai

基本 RAG 管道

python
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Settings
from llama_index.embeddings.openai import OpenAIEmbedding
from llama_index.llms.openai import OpenAI
配置模型
Settings.llm = OpenAI(model='gpt-4o-mini', temperature=0.1)
Settings.embed_model = OpenAIEmbedding(model='text-embedding-3-small')
加载文档
documents = SimpleDirectoryReader('./docs').load_data()
构建索引（自动嵌入和存储）
index = VectorStoreIndex.from_documents(documents)
查询
query_engine = index.as_query_engine(similarity_top_k=5)
response = query_engine.query('退款政策是什么？')
print(response.response)
访问源文档
for node in response.source_nodes:
    print(f'得分: {node.score:.3f} | {node.text[:100]}')

使用 Qdrant 持久化存储

python
from llama_index.vector_stores.qdrant import QdrantVectorStore
from llama_index.core import StorageContext
from qdrant_client import QdrantClient
client = QdrantClient(url='http://localhost:6333')
vector_store = QdrantVectorStore(client=client, collection_name='docs')
storage_context = StorageContext.from_defaults(vector_store=vector_store)
构建并持久化
index = VectorStoreIndex.from_documents(
    documents,
    storage_context=storage_context
)
加载已有索引（无需重新嵌入）
index = VectorStoreIndex.from_vector_store(vector_store)
query_engine = index.as_query_engine()

高级查询模式

python
聊天引擎（保持对话）
chat_engine = index.as_chat_engine(chat_mode='condense_plus_context')
response = chat_engine.chat('退款政策是什么？')
print(response.response)
response = chat_engine.chat('处理需要多长时间？')  # 记住上下文
print(response.response)
子问题引擎（分解复杂查询）
from llama_index.core.query_engine import SubQuestionQueryEngine
from llama_index.core.tools import QueryEngineTool, ToolMetadata
tools = [
    QueryEngineTool(
        query_engine=query_engine,
        metadata=ToolMetadata(
            name='docs',
            description='公司文档和政策'
        )
    )
]
sub_question_engine = SubQuestionQueryEngine.from_defaults(query_engine_tools=tools)
response = sub_question_engine.query(
    '比较我们的退款政策和退货政策，告诉我哪个对客户更友好'
)
print(response.response)

文档摄取管道

python
from llama_index.core.ingestion import IngestionPipeline
from llama_index.core.node_parser import SentenceSplitter, SemanticSplitterNodeParser
from llama_index.core.extractors import TitleExtractor, SummaryExtractor
构建带有转换的摄取管道
pipeline = IngestionPipeline(
    transformations=[
        # 分割为语义块
        SemanticSplitterNodeParser(buffer_size=1, embed_model=Settings.embed_model),
        # 提取元数据
        TitleExtractor(nodes=5),
        SummaryExtractor(summaries=['prev', 'self']),
        # 嵌入并存储
        Settings.embed_model
    ],
    vector_store=vector_store  # 自动存储嵌入
)
nodes = pipeline.run(documents=documents)
print(f'创建了 {len(nodes)} 个节点并带有嵌入')

元数据过滤

python
from llama_index.core.vector_stores import MetadataFilter, MetadataFilters
按文档元数据过滤
filters = MetadataFilters(filters=[
    MetadataFilter(key='department', value='legal'),
    MetadataFilter(key='year', value='2026')
])
query_engine = index.as_query_engine(
    similarity_top_k=5,
    filters=filters
)
response = query_engine.query('我们的合规要求是什么？')

流式响应

python
streaming_engine = index.as_query_engine(streaming=True)
streaming_response = streaming_engine.query('解释我们的数据保留政策')
streaming_response.print_response_stream()  # 流式输出到标准输出
或者手动迭代
for token in streaming_response.response_gen:
    print(token, end='', flush=True)

结论

LlamaIndex 是 2026 年生产级 RAG 最全面的框架。其摄取管道、多种查询模式和元数据过滤使其成为企业文档问答应用的理想选择。