Master LangChain Chains with LCEL: From Simple Jokes to RAG and Agent Pipelines

In LangChain, a Chain links components such as language models, prompts, and tools in sequence to accomplish a task. By chaining simple components, complex applications can be built.

LCEL: LangChain Expression Language

LCEL is the declarative method introduced in LangChain v0.1.0. It uses the pipe symbol | to connect components, making data flow explicit and providing built‑in support for streaming, parallel execution, and logging.

A minimal chain consists of three parts:

Prompt : receives user input and formats it into a model‑understandable prompt.

Model : consumes the formatted prompt and generates a response.

Output Parser : transforms the model’s textual response into a usable structure (e.g., JSON or a Python object).

Running the Example Repository

The 22‑Chains/ directory contains several scripts that demonstrate LCEL chains of increasing complexity.

Environment Setup

API keys : create a .env file in 22‑Chains/ and add your keys.

OPENAI_API_KEY="sk-..."
TAVILY_API_KEY="tvly-..."

Obtain a free Tavily AI key at https://tavily.com/.

Install dependencies : run the following command once to install all required packages.

pip install langchain langchain-openai langchain-community python-dotenv faiss-cpu tiktoken tavily-python

Execute an example : run any script from the terminal.

python example_1_simple_lcel_chain.py
python example_2_rag_chain.py
python example_3_chain_with_history.py
python example_1_basic_agent.py
python example_2_agent_with_memory.py

Example File Walk‑through

example_1_simple_lcel_chain.py

: builds the simplest LCEL chain—prompt → model → output parser—to generate a joke. example_2_rag_chain.py: constructs a Retrieval‑Augmented Generation (RAG) chain. First it queries a vector store for relevant documents, then feeds those documents to the model so the answer is grounded in retrieved context. example_3_chain_with_history.py: demonstrates a conversational chain with manual memory management. The script appends each turn to a chat history list, allowing the model to retain context across multiple exchanges. example_1_basic_agent.py and example_2_agent_with_memory.py: introduce agents, a dynamic chain type. An agent uses a large language model to decide which tool (e.g., a web search via Tavily) to invoke. The second script adds a memory component so the agent can recall prior interactions.

Note: These examples illustrate the full spectrum of LangChain chaining—from basic LCEL composition to advanced RAG and agent patterns.

Reference

[1] Tavily AI: https://tavily.com/