LangChain - AI Orchestration

LangChain is a framework for building applications with LLMs - chains of operations, memory, tools!

Installation

1pip install langchain langchain-openai langchain-anthropic

LangChain Basics

Chat Models

1from langchain_openai import ChatOpenAI
2from langchain_anthropic import ChatAnthropic
3from langchain_core.messages import HumanMessage, SystemMessage
4
5# OpenAI
6llm = ChatOpenAI(model="gpt-5", temperature=0.7)
7
8# Anthropic
9llm = ChatAnthropic(model="claude-sonnet-4-6")
10
11# Simple invocation
12messages = [
13    SystemMessage(content="You are a Safari expert."),
14    HumanMessage(content="Tell me about elephants.")
15]
16
17response = llm.invoke(messages)
18print(response.content)

Prompt Templates

1from langchain_core.prompts import ChatPromptTemplate
2
3# Template with variables
4prompt = ChatPromptTemplate.from_messages([
5    ("system", "You are an expert in {field}."),
6    ("human", "Tell me about {topic}.")
7])
8
9# Formatting
10formatted = prompt.format_messages(
11    field="wildlife",
12    topic="wildebeest migration"
13)
14
15response = llm.invoke(formatted)

Chains - Operation Pipelines

1from langchain_core.output_parsers import StrOutputParser
2
3# Simple chain: prompt -> model -> parser
4chain = prompt | llm | StrOutputParser()
5
6# Run
7result = chain.invoke({
8    "field": "Safari",
9    "topic": "lions"
10})
11print(result)

Structured Output

1from langchain_core.pydantic_v1 import BaseModel, Field
2from langchain_core.output_parsers import JsonOutputParser
3
4# Structure definition
5class AnimalInfo(BaseModel):
6    name: str = Field(description="Animal name")
7    habitat: str = Field(description="Habitat")
8    diet: str = Field(description="Diet")
9    population: int = Field(description="Population")
10
11# Parser
12parser = JsonOutputParser(pydantic_object=AnimalInfo)
13
14# Prompt with format instruction
15prompt = ChatPromptTemplate.from_messages([
16    ("system", "Return information about the animal in JSON format."),
17    ("human", "{format_instructions}\n\nAnimal: {animal}")
18])
19
20chain = prompt | llm | parser
21
22result = chain.invoke({
23    "animal": "African elephant",
24    "format_instructions": parser.get_format_instructions()
25})
26print(result)  # {'name': 'African elephant', ...}

Memory - Conversation History

1from langchain_community.chat_message_histories import ChatMessageHistory
2from langchain_core.runnables.history import RunnableWithMessageHistory
3
4# Message history
5history = ChatMessageHistory()
6
7# Chain with memory
8chain_with_history = RunnableWithMessageHistory(
9    chain,
10    lambda session_id: history,
11    input_messages_key="input",
12    history_messages_key="history"
13)
14
15# First message
16response1 = chain_with_history.invoke(
17    {"input": "I'm on a Safari in Kenya"},
18    config={"configurable": {"session_id": "abc123"}}
19)
20
21# Second message - the model remembers context
22response2 = chain_with_history.invoke(
23    {"input": "What animals can I see here?"},
24    config={"configurable": {"session_id": "abc123"}}
25)