The Beginner’s Guide to Using LangChain to Build With LLMs

Large language models (LLMs) are incredibly powerful for their reasoning capabilities, but working with them presents challenges that are different from building traditional software:

• Calls tend to be much slower, but stream generated output as it becomes available.
• They take unstructured natural language as input, and are capable of “understanding” subtleties of that language.
• They are non-deterministic - you may get different outputs even with the same inputs.

LangChain is a popular framework for developing LLM-powered apps built with these challenges in mind. It also provides a wide range of integrations with closed-source model providers (like OpenAI, Anthropic, and Google), open source models, and other third-party components like vectorstores.

This article will walk through the fundamentals of building with LLMs and LangChain’s Python library. It will only assume you have basic familiarity with Python, and doesn’t require any ML background.

You’ll learn about:

• Some of the basic components of LangChain.
• LangChain Expression Language, the protocol that LangChain is built on and what underpins component chaining and streaming.
• Using outside data to steer the model’s output.
• Debugging and tracing the internals of your chains.

Let’s dive in!

Setup

We recommend using a Jupyter notebook to run the code in this tutorial since it provides a clean, interactive environment. See this page for instructions on setting it up locally, or check out Google Colab for an in-browser experience.

Let’s install the required dependencies. This guide defaults to Anthropic and their Claude 3 Chat Models, but you can use LangChain with a wide range of options:

pip install langchain_core langchain_anthropic

<aside> 💡 If you’re working in a Jupyter notebook, you’ll need to prefix pip with a % symbol like this: %pip install langchain_core langchain_anthropic

</aside>

You’ll also need an Anthropic API key, which you can obtain here from their official console. Once you have it, set as an environment variable named ANTHROPIC_API_KEY:

export ANTHROPIC_API_KEY="..."