LangChain vs LlamaIndex vs Haystack：2026年RAG框架对比

构建生产级RAG系统需要选择合适的框架。每个框架在文档索引、检索和LLM编排方面都有不同的方法。

框架理念

LangChain：可组合的链和代理，拥有200+集成的广泛生态系统

LlamaIndex：以数据为中心，针对文档索引和复杂查询优化

Haystack：面向生产，模块化管道架构，支持MLOps

LangChain RAG管道

python
from langchain.document_loaders import DirectoryLoader, PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_community.vectorstores import Chroma
from langchain_core.runnables import RunnablePassthrough
from langchain import hub
加载并分割文档
loader = DirectoryLoader('./docs', glob='**/*.pdf', loader_cls=PyPDFLoader)
chunks = RecursiveCharacterTextSplitter(
    chunk_size=1000, chunk_overlap=200
).split_documents(loader.load())
创建向量存储
vectorstore = Chroma.from_documents(
    chunks,
    OpenAIEmbeddings(model='text-embedding-3-large'),
    persist_directory='./chroma_db'
)
创建RAG链
rag_chain = (
    {'context': vectorstore.as_retriever(search_kwargs={'k': 5}),
     'question': RunnablePassthrough()}
    | hub.pull('rlm/rag-prompt')
    | ChatOpenAI(model='gpt-5', temperature=0)
)
response = rag_chain.invoke('关键发现是什么？')
print(response.content)

LlamaIndex RAG管道

python
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Settings
from llama_index.embeddings.openai import OpenAIEmbedding
from llama_index.llms.anthropic import Anthropic
from llama_index.core.node_parser import SentenceSplitter
from llama_index.core.postprocessor import SimilarityPostprocessor
from llama_index.core.query_engine import RetrieverQueryEngine
from llama_index.core.retrievers import VectorIndexRetriever
配置设置
Settings.llm = Anthropic(model='claude-sonnet-4-5')
Settings.embed_model = OpenAIEmbedding(model='text-embedding-3-large')
Settings.node_parser = SentenceSplitter(chunk_size=512, chunk_overlap=20)
构建索引
documents = SimpleDirectoryReader('./docs').load_data()
index = VectorStoreIndex.from_documents(documents, show_progress=True)
带后处理的高级查询
query_engine = RetrieverQueryEngine(
    retriever=VectorIndexRetriever(index=index, similarity_top_k=10),
    node_postprocessors=[SimilarityPostprocessor(similarity_cutoff=0.7)]
)
response = query_engine.query('方法论是什么？')
print(response)
print(f'来源数量：{len(response.source_nodes)}')

Haystack RAG管道

python
from haystack import Pipeline
from haystack.components.converters import PyPDFToDocument
from haystack.components.preprocessors import DocumentSplitter
from haystack.components.embedders import OpenAIDocumentEmbedder, OpenAITextEmbedder
from haystack.components.writers import DocumentWriter
from haystack.components.retrievers.in_memory import InMemoryEmbeddingRetriever
from haystack.document_stores.in_memory import InMemoryDocumentStore
from haystack.components.builders import PromptBuilder
from haystack.components.generators import OpenAIGenerator
doc_store = InMemoryDocumentStore()
索引管道
indexing = Pipeline()
indexing.add_component('converter', PyPDFToDocument())
indexing.add_component('splitter', DocumentSplitter(split_by='word', split_length=200))
indexing.add_component('embedder', OpenAIDocumentEmbedder(model='text-embedding-3-large'))
indexing.add_component('writer', DocumentWriter(document_store=doc_store))
indexing.connect('converter', 'splitter')
indexing.connect('splitter', 'embedder')
indexing.connect('embedder', 'writer')
indexing.run({'converter': {'sources': ['./docs/report.pdf']}})
查询管道
template = '根据上下文回答：{% for doc in documents %}{{ doc.content }}{% endfor %}\n问题：{{question}}'
querying = Pipeline()
querying.add_component('embedder', OpenAITextEmbedder(model='text-embedding-3-large'))
querying.add_component('retriever', InMemoryEmbeddingRetriever(document_store=doc_store, top_k=5))
querying.add_component('prompt', PromptBuilder(template=template))
querying.add_component('llm', OpenAIGenerator(model='gpt-5'))
querying.connect('embedder.embedding', 'retriever.query_embedding')
querying.connect('retriever', 'prompt.documents')
querying.connect('prompt', 'llm')
result = querying.run({'embedder': {'text': '关键发现'}, 'prompt': {'question': '关键发现'}})

性能对比

指标LangChainLlamaIndexHaystack

索引1000个文档45秒38秒42秒 查询延迟1.8秒1.4秒1.6秒 内存使用高中低 生产就绪度⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐⭐ 社区规模最大大中

决策指南

LangChain：复杂的代理链，最大的生态系统，工具集成

LlamaIndex：文档问答，高级索引策略，数据管道

Haystack：生产部署，MLOps，搜索型应用

结论

LlamaIndex在纯RAG性能和开发者体验方面胜出。LangChain在生态系统广度上占优。Haystack在生产稳健性上领先。对于2026年的新RAG项目，建议从LlamaIndex开始——其文档理解能力无与伦比。