langchain-framework

What this skill does

# LangChain Framework

---
progressive_disclosure:
  entry_point:
    summary: "LLM application framework with chains, agents, RAG, and memory"
    when_to_use:
      - "When building LLM-powered applications"
      - "When implementing RAG (Retrieval Augmented Generation)"
      - "When creating AI agents with tools"
      - "When chaining multiple LLM calls"
    quick_start:
      - "pip install langchain langchain-anthropic"
      - "Set up LLM (ChatAnthropic or ChatOpenAI)"
      - "Create chain with prompts and LLM"
      - "Invoke chain with input"
  token_estimate:
    entry: 85
    full: 5200
---

## Core Concepts

### LangChain Expression Language (LCEL)
Modern composable syntax for building chains with `|` operator.

**Basic Chain**:
```python
from langchain_anthropic import ChatAnthropic
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser

# Components
llm = ChatAnthropic(model="claude-3-5-sonnet-20241022")
prompt = ChatPromptTemplate.from_template("Tell me a joke about {topic}")
output_parser = StrOutputParser()

# Compose with LCEL
chain = prompt | llm | output_parser

# Invoke
result = chain.invoke({"topic": "programming"})
```

**Why LCEL**:
- Type safety and auto-completion
- Streaming support built-in
- Async by default
- Observability with LangSmith
- Easier debugging

### Chain Components

**Prompts**:
```python
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder

# Simple template
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant."),
    ("user", "{input}")
])

# With message history
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant."),
    MessagesPlaceholder(variable_name="history"),
    ("user", "{input}")
])

# Few-shot examples
from langchain_core.prompts import FewShotChatMessagePromptTemplate

examples = [
    {"input": "2+2", "output": "4"},
    {"input": "3*5", "output": "15"}
]

example_prompt = ChatPromptTemplate.from_messages([
    ("human", "{input}"),
    ("ai", "{output}")
])

few_shot_prompt = FewShotChatMessagePromptTemplate(
    example_prompt=example_prompt,
    examples=examples
)
```

**LLMs**:
```python
# Anthropic Claude
from langchain_anthropic import ChatAnthropic

llm = ChatAnthropic(
    model="claude-3-5-sonnet-20241022",
    temperature=0.7,
    max_tokens=1024,
    timeout=60.0
)

# OpenAI
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="gpt-4-turbo-preview",
    temperature=0.7
)

# Streaming
for chunk in llm.stream("Tell me a story"):
    print(chunk.content, end="", flush=True)
```

**Output Parsers**:
```python
from langchain_core.output_parsers import StrOutputParser, JsonOutputParser
from langchain.output_parsers import PydanticOutputParser
from pydantic import BaseModel, Field

# String parser
str_parser = StrOutputParser()

# JSON parser
json_parser = JsonOutputParser()

# Structured output
class Person(BaseModel):
    name: str = Field(description="Person's name")
    age: int = Field(description="Person's age")

parser = PydanticOutputParser(pydantic_object=Person)
prompt = ChatPromptTemplate.from_template(
    "Extract person info.\n{format_instructions}\n{query}"
)
chain = prompt | llm | parser
```

## RAG (Retrieval Augmented Generation)

### Document Loading
```python
from langchain_community.document_loaders import (
    TextLoader,
    PyPDFLoader,
    DirectoryLoader,
    WebBaseLoader
)

# Text files
loader = TextLoader("document.txt")
docs = loader.load()

# PDFs
loader = PyPDFLoader("document.pdf")
docs = loader.load()

# Directory of files
loader = DirectoryLoader(
    "./docs",
    glob="**/*.md",
    show_progress=True
)
docs = loader.load()

# Web pages
loader = WebBaseLoader("https://example.com")
docs = loader.load()
```

### Text Splitting
```python
from langchain.text_splitter import (
    RecursiveCharacterTextSplitter,
    CharacterTextSplitter,
    TokenTextSplitter
)

# Recursive splitter (recommended)
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=1000,
    chunk_overlap=200,
    length_function=len,
    separators=["\n\n", "\n", " ", ""]
)

chunks = text_splitter.split_documents(docs)

# Token-aware splitting
from langchain.text_splitter import TokenTextSplitter

splitter = TokenTextSplitter(
    chunk_size=512,
    chunk_overlap=50
)
```

### Vector Stores
```python
from langchain_community.vectorstores import Chroma, FAISS, Pinecone
from langchain_openai import OpenAIEmbeddings
from langchain_community.embeddings import HuggingFaceEmbeddings

# Embeddings
embeddings = OpenAIEmbeddings()

# Chroma (local, persistent)
vectorstore = Chroma.from_documents(
    documents=chunks,
    embedding=embeddings,
    persist_directory="./chroma_db"
)

# FAISS (local, in-memory)
vectorstore = FAISS.from_documents(
    documents=chunks,
    embedding=embeddings
)
vectorstore.save_local("./faiss_index")

# Pinecone (cloud)
from langchain_community.vectorstores import Pinecone
import pinecone

pinecone.init(api_key="your-key", environment="us-west1-gcp")
vectorstore = Pinecone.from_documents(
    documents=chunks,
    embedding=embeddings,
    index_name="langchain-index"
)
```

### RAG Chain
```python
from langchain_core.runnables import RunnablePassthrough
from langchain_core.output_parsers import StrOutputParser

# Create retriever
retriever = vectorstore.as_retriever(
    search_type="similarity",
    search_kwargs={"k": 4}
)

# RAG prompt
template = """Answer based on context:

Context: {context}

Question: {question}

Answer:"""

prompt = ChatPromptTemplate.from_template(template)

# Format documents
def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)

# RAG chain
rag_chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

# Query
answer = rag_chain.invoke("What is LangChain?")
```

### Advanced RAG Patterns
```python
# Multi-query retrieval
from langchain.retrievers.multi_query import MultiQueryRetriever

retriever = MultiQueryRetriever.from_llm(
    retriever=vectorstore.as_retriever(),
    llm=llm
)

# Contextual compression
from langchain.retrievers import ContextualCompressionRetriever
from langchain.retrievers.document_compressors import LLMChainExtractor

compressor = LLMChainExtractor.from_llm(llm)
compression_retriever = ContextualCompressionRetriever(
    base_compressor=compressor,
    base_retriever=vectorstore.as_retriever()
)

# Parent document retriever
from langchain.retrievers import ParentDocumentRetriever
from langchain.storage import InMemoryStore

store = InMemoryStore()
retriever = ParentDocumentRetriever(
    vectorstore=vectorstore,
    docstore=store,
    child_splitter=text_splitter
)
```

## Agents and Tools

### Tool Creation
```python
from langchain.tools import tool
from langchain_core.tools import Tool

# Decorator approach
@tool
def search_wikipedia(query: str) -> str:
    """Search Wikipedia for information."""
    # Implementation
    return f"Results for: {query}"

# Class approach
from langchain.tools import BaseTool
from pydantic import BaseModel, Field

class CalculatorInput(BaseModel):
    expression: str = Field(description="Mathematical expression")

class CalculatorTool(BaseTool):
    name = "calculator"
    description = "Useful for math calculations"
    args_schema = CalculatorInput

    def _run(self, expression: str) -> str:
        return str(eval(expression))

# Pre-built tools
from langchain_community.tools import (
    DuckDuckGoSearchRun,
    WikipediaQueryRun,
    PythonREPLTool
)

search = DuckDuckGoSearchRun()
wikipedia = WikipediaQueryRun()
python_repl = PythonREPLTool()
```

### Agent Types
```python
from langchain.agents import create_react_agent, AgentExecutor
from langchain import hub

# ReAct agent (recommended)
prompt = hub.pull("hwchase17/react")
tools = [search_wikipedia, CalculatorTool()]

agent = create_react_age