Getting Started with LangChain: Building LLM Applications in Python

Since the release of ChatGPT, large language models (LLMs) have become popular. This article introduces LangChain, an open‑source Python framework that simplifies building LLM‑powered applications.

What is LangChain? LangChain provides a unified interface to many base models, prompt management, memory, and tool integration.

Prerequisites Python ≥ 3.8.1, pip, and API keys for the models you intend to use.

Installation

pip install langchain

import langchain

API keys – obtain keys for OpenAI, Hugging Face, or other providers and set them in environment variables.

import os
os.environ["OPENAI_API_KEY"] = "..."  # insert your API token

os.environ["HUGGINGFACEHUB_API_TOKEN"] = "..."  # insert your API token

Core modules

Model – load proprietary or open‑source LLMs and embedding models.

Prompt – create PromptTemplate and FewShotPromptTemplate for prompt engineering.

Chain – combine LLMs with prompts, sequential chains, or retrieval QA.

Index – load external data (e.g., YouTube transcripts) and store embeddings in a vector store such as FAISS.

Memory – use ConversationChain to retain chat history.

Agent – enable LLMs to call external tools like Wikipedia or a calculator.

Example code snippets illustrate each component, e.g., creating an LLMChain, a SimpleSequentialChain, a RetrievalQA chain, and an agent that queries Wikipedia and performs arithmetic.

from langchain.chains import LLMChain, SimpleSequentialChain
chain = LLMChain(llm=llm, prompt=prompt)
overall_chain = SimpleSequentialChain(chains=[chain, chain_two], verbose=True)

from langchain.agents import load_tools, initialize_agent, AgentType
tools = load_tools(["wikipedia", "llm-math"], llm=llm)
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)

Summary – LangChain enables rapid prototyping of LLM applications on a personal laptop, offering interfaces for models, prompts, memory, retrieval, and tool use. The library is evolving quickly, so examples may become outdated.