LangChain - Construa aplicações com LLM, agentes e RAG

O framework mais popular para construir aplicações powered por LLM.

Quando usar LangChain

Use LangChain quando:

• Construir agentes com tool calling e reasoning (padrão ReAct)
• Implementar pipelines RAG (retrieval-augmented generation)
• Precisar trocar provedores de LLM facilmente (OpenAI, Anthropic, Google)
• Criar chatbots com memória de conversa
• Prototipagem rápida de aplicações com LLM
• Deployments em produção com observabilidade LangSmith

Métricas:

• 119.000+ stars no GitHub
• 272.000+ repositórios usam LangChain
• 500+ integrações (modelos, vector stores, tools)
• 3.800+ contribuidores

Use alternativas em vez disso:

• LlamaIndex: focado em RAG, melhor para Q&A de documentos
• LangGraph: workflows stateful complexos, mais controle
• Haystack: pipelines de busca em produção
• Semantic Kernel: ecossistema Microsoft

Início rápido

Instalação

# Biblioteca principal (Python 3.10+)
pip install -U langchain

# Com OpenAI
pip install langchain-openai

# Com Anthropic
pip install langchain-anthropic

# Extras comuns
pip install langchain-community  # 500+ integrações
pip install langchain-chroma     # Vector store

Uso básico de LLM

from langchain_anthropic import ChatAnthropic

# Inicializar modelo
llm = ChatAnthropic(model="claude-sonnet-4-5-20250929")

# Conclusão simples
response = llm.invoke("Explain quantum computing in 2 sentences")
print(response.content)

Criar um agente (padrão ReAct)

from langchain.agents import create_agent
from langchain_anthropic import ChatAnthropic

# Definir tools
def get_weather(city: str) -> str:
    """Get current weather for a city."""
    return f"It's sunny in {city}, 72°F"

def search_web(query: str) -> str:
    """Search the web for information."""
    return f"Search results for: {query}"

# Criar agente (<10 linhas!)
agent = create_agent(
    model=ChatAnthropic(model="claude-sonnet-4-5-20250929"),
    tools=[get_weather, search_web],
    system_prompt="You are a helpful assistant. Use tools when needed."
)

# Executar agente
result = agent.invoke({"messages": [{"role": "user", "content": "What's the weather in Paris?"}]})
print(result["messages"][-1].content)

Conceitos principais

1. Models - abstração de LLM

from langchain_openai import ChatOpenAI
from langchain_anthropic import ChatAnthropic
from langchain_google_genai import ChatGoogleGenerativeAI

# Trocar provedores facilmente
llm = ChatOpenAI(model="gpt-4o")
llm = ChatAnthropic(model="claude-sonnet-4-5-20250929")
llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash-exp")

# Streaming
for chunk in llm.stream("Write a poem"):
    print(chunk.content, end="", flush=True)

2. Chains - operações sequenciais

from langchain.chains import LLMChain
from langchain.prompts import PromptTemplate

# Definir template de prompt
prompt = PromptTemplate(
    input_variables=["topic"],
    template="Write a 3-sentence summary about {topic}"
)

# Criar chain
chain = LLMChain(llm=llm, prompt=prompt)

# Executar chain
result = chain.run(topic="machine learning")

3. Agents - reasoning com tools

Padrão ReAct (Reasoning + Acting):

from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain.tools import Tool

# Definir tool customizada
calculator = Tool(
    name="Calculator",
    func=lambda x: eval(x),
    description="Useful for math calculations. Input: valid Python expression."
)

# Criar agente com tools
agent = create_tool_calling_agent(
    llm=llm,
    tools=[calculator, search_web],
    prompt="Answer questions using available tools"
)

# Criar executor
agent_executor = AgentExecutor(agent=agent, tools=[calculator], verbose=True)

# Executar com reasoning
result = agent_executor.invoke({"input": "What is 25 * 17 + 142?"})

4. Memory - histórico de conversa

from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationChain

# Adicionar memória para rastrear conversa
memory = ConversationBufferMemory()

conversation = ConversationChain(
    llm=llm,
    memory=memory,
    verbose=True
)

# Conversa multi-turno
conversation.predict(input="Hi, I'm Alice")
conversation.predict(input="What's my name?")  # Remembers "Alice"

RAG (Retrieval-Augmented Generation)

Pipeline RAG básico

from langchain_community.document_loaders import WebBaseLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings
from langchain_chroma import Chroma
from langchain.chains import RetrievalQA

# 1. Carregar documentos
loader = WebBaseLoader("https://docs.python.org/3/tutorial/")
docs = loader.load()

# 2. Dividir em chunks
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=1000,
    chunk_overlap=200
)
splits = text_splitter.split_documents(docs)

# 3. Criar embeddings e vector store
vectorstore = Chroma.from_documents(
    documents=splits,
    embedding=OpenAIEmbeddings()
)

# 4. Criar retriever
retriever = vectorstore.as_retriever(search_kwargs={"k": 4})

# 5. Criar chain de QA
qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    retriever=retriever,
    return_source_documents=True
)

# 6. Consultar
result = qa_chain({"query": "What are Python decorators?"})
print(result["result"])
print(f"Sources: {result['source_documents']}")

RAG conversacional com memória

from langchain.chains import ConversationalRetrievalChain

# RAG com memória de conversa
qa = ConversationalRetrievalChain.from_llm(
    llm=llm,
    retriever=retriever,
    memory=ConversationBufferMemory(
        memory_key="chat_history",
        return_messages=True
    )
)

# RAG multi-turno
qa({"question": "What is Python used for?"})
qa({"question": "Can you elaborate on web development?"})  # Remembers context

Padrões avançados de agentes

Saída estruturada

from langchain_core.pydantic_v1 import BaseModel, Field

# Definir schema
class WeatherReport(BaseModel):
    city: str = Field(description="City name")
    temperature: float = Field(description="Temperature in Fahrenheit")
    condition: str = Field(description="Weather condition")

# Obter resposta estruturada
structured_llm = llm.with_structured_output(WeatherReport)
result = structured_llm.invoke("What's the weather in SF? It's 65F and sunny")
print(result.city, result.temperature, result.condition)

Execução paralela de tools

from langchain.agents import create_tool_calling_agent

# Agente paraleliza automaticamente tool calls independentes
agent = create_tool_calling_agent(
    llm=llm,
    tools=[get_weather, search_web, calculator]
)

# Isso chamará get_weather("Paris") e get_weather("London") em paralelo
result = agent.invoke({
    "messages": [{"role": "user", "content": "Compare weather in Paris and London"}]
})

Streaming de execução de agente

# Fazer stream dos passos do agente
for step in agent_executor.stream({"input": "Research AI trends"}):
    if "actions" in step:
        print(f"Tool: {step['actions'][0].tool}")
    if "output" in step:
        print(f"Output: {step['output']}")

Padrões comuns

QA multi-documento

from langchain.chains.qa_with_sources import load_qa_with_sources_chain

# Carregar múltiplos documentos
docs = [
    loader.load("https://docs.python.org"),
    loader.load("https://docs.numpy.org")
]

# QA com citações de fontes
chain = load_qa_with_sources_chain(llm, chain_type="stuff")
result = chain({"input_documents": docs, "question": "How to use numpy arrays?"})
print(result["output_text"])  # Inclui citações de fontes

Tools customizadas com tratamento de erros

from langchain.tools import tool

@tool
def risky_operation(query: str) -> str:
    """Perform a risky operation that might fail."""
    try:
        # Your operation here
        result = perform_operation(query)
        return f"Success: {result}"
    except Exception as e:
        return f"Error: {str(e)}"

# Agente trata erros com graça
agent = create_agent(model=llm, tools=[risky_operation])

Observabilidade LangSmith

import os

# Habilitar tracing
os.environ["LANGCHAIN_TRACING_V2"] = "true"
os.environ["LANGCHAIN_API_KEY"] = "your-api-key"
os.environ["LANGCHAIN_PROJECT"] = "my-project"

# Todos os chains/agentes são rastreados automaticamente
agent = create_agent(model=llm, tools=[calculator])
result = agent.invoke({"input": "Calculate 123 * 456"})

# Ver traces em smith.langchain.com

Vector stores

Chroma (local)

from langchain_chroma import Chroma

vectorstore = Chroma.from_documents(
    documents=docs,
    embedding=OpenAIEmbeddings(),
    persist_directory="./chroma_db"
)

Pinecone (cloud)

from langchain_pinecone import PineconeVectorStore

vectorstore = PineconeVectorStore.from_documents(
    documents=docs,
    embedding=OpenAIEmbeddings(),
    index_name="my-index"
)

FAISS (busca de similaridade)

from langchain_community.vectorstores import FAISS

vectorstore = FAISS.from_documents(docs, OpenAIEmbeddings())
vectorstore.save_local("faiss_index")

# Carregar depois
vectorstore = FAISS.load_local("faiss_index", OpenAIEmbeddings())

Document loaders

# Páginas web
from langchain_community.document_loaders import WebBaseLoader
loader = WebBaseLoader("https://example.com")

# PDFs
from langchain_community.document_loaders import PyPDFLoader
loader = PyPDFLoader("paper.pdf")

# GitHub
from langchain_community.document_loaders import GithubFileLoader
loader = GithubFileLoader(repo="user/repo", file_filter=lambda x: x.endswith(".py"))

# CSV
from langchain_community.document_loaders import CSVLoader
loader = CSVLoader("data.csv")

Text splitters

# Recursivo (recomendado para texto geral)
from langchain.text_splitter import RecursiveCharacterTextSplitter
splitter = RecursiveCharacterTextSplitter(
    chunk_size=1000,
    chunk_overlap=200,
    separators=["\n\n", "\n", " ", ""]
)

# Code-aware
from langchain.text_splitter import PythonCodeTextSplitter
splitter = PythonCodeTextSplitter(chunk_size=500)

# Semântico (por significado)
from langchain_experimental.text_splitter import SemanticChunker
splitter = SemanticChunker(OpenAIEmbeddings())

Melhores práticas

1. Comece simples - Use create_agent() na maioria dos casos
2. Habilite streaming - Melhor UX para respostas longas
3. Adicione tratamento de erros - Tools podem falhar, trate com graça
4. Use LangSmith - Essencial para debugar agentes
5. Otimize tamanho de chunk - 500-1000 caracteres para RAG
6. Versionize prompts - Rastreie mudanças em produção
7. Cache de embeddings - Caro, cache quando possível
8. Monitore custos - Rastreie uso de tokens com LangSmith

Benchmarks de performance

Operação	Latência	Notas
Simple LLM call	~1-2s	Depende do provedor
Agent with 1 tool	~3-5s	Overhead de reasoning ReAct
RAG retrieval	~0.5-1s	Vector search + LLM
Embedding 1000 docs	~10-30s	Depende do modelo

LangChain vs LangGraph

Feature	LangChain	LangGraph
Ideal para	Quick agents, RAG	Workflows complexos
Nível de abstração	Alto	Baixo
Código para começar	<10 linhas	~30 linhas
Controle	Simples	Total
Workflows stateful	Limitado	Nativo
Grafos cíclicos	Não	Sim
Human-in-loop	Básico	Avançado

Use LangGraph quando:

• Precisar de workflows stateful com ciclos
• Exigir controle refinado
• Construir sistemas multi-agente
• Aplicações em produção com lógica complexa

Referências

• Agents Guide - ReAct, tool calling, streaming
• RAG Guide - Document loaders, retrievers, QA chains
• Integration Guide - Vector stores, LangSmith, deployment

Recursos

• GitHub: https://github.com/langchain-ai/langchain ⭐ 119.000+
• Docs: https://docs.langchain.com
• API Reference: https://reference.langchain.com/python
• LangSmith: https://smith.langchain.com (observabilidade)
• Version: 0.3+ (estável)
• License: MIT