What are Chains? The "Lang" in LangChain

In the previous modules, we handled Prompts and Models as separate variables.

1. text = prompt.format(topic="...")
2. response = model.invoke(text)

This is fine for one step. But what if you have 10 steps? What if you want to add a logging step, a parsing step, and a database step? Manual management becomes a mess. Chains allow you to "Pipe" these steps together into a single, reusable object.

1. The Unix Philosophy

LangChain follows the Unix "Pipe" philosophy: Input | Step1 | Step2 | Output. The output of the Prompt becomes the input of the Model. The output of the Model becomes the input of the Parser.

---

2. Introducing LCEL (LangChain Expression Language)

LCEL is the modern way to build chains. It uses the | (pipe) operator. It is significantly faster and more readable than the older, legacy LLMChain classes.

# The LCEL Way
chain = prompt | model | parser

---

3. Visualizing the Pipeline

graph LR
    Input[User Variable] --> P[Prompt Template]
    P -->|Formatted String| M[LLM Model]
    M -->|AIMessage| O[Output Parser]
    O --> Final[Clean Python Object]

---

4. Why use Chains instead of simple Python calls?

1. Streaming: If you pipe a prompt to a model, the chain automatically supports streaming out of the box.
2. Async: Chains are natively asynchronous.
3. Observability: Tools like LangSmith can "see" inside the pipe and show you where it broke.
4. Parallelism: Chains can automatically run independent steps in parallel.

---

5. The "Standard" Chain

The most common chain you will ever build consists of three parts:

1. Prompt: Converts variables to a string.
2. Model: Converts a string to an AI response.
3. StrOutputParser: Converts an AI response into a simple string.

---

Key Takeaways

• Chains are the fundamental unit of work in LangChain.
• LCEL is the language used to build these chains using the | operator.
• Chains automate the "Hand-off" between different AI components.
• Using chains enables Streaming, Async, and Tracing automatically.