Overview

LangChain Expression Language (LCEL) is a declarative chain composition syntax that uses the pipe operator | to chain Runnable components together, enabling complex workflows. LCEL is the recommended way to compose chains in LangChain v0.1+.

Basic Syntax

Pipe Operator |

from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import StrOutputParser

model = ChatOpenAI(model="gpt-4")
parser = StrOutputParser()

chain = model | parser
result = chain.invoke("What is LCEL?")

Runnable Interface

All LCEL-compatible components implement Runnable:

chain.invoke(input)           # Sync
chain.batch([input1, input2]) # Batch
chain.stream("prompt")        # Stream

Common Components

Prompt Templates

from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a {language} assistant"),
    ("human", "{question}")
])

chain = prompt | model | parser

Output Parsers

from langchain_core.output_parsers import JsonOutputParser

parser = JsonOutputParser()
chain = prompt | model | parser

Composition Patterns

Parallel Processing

from langchain_core.runnables import RunnableParallel

combined = RunnableParallel(
    detail=chain1,
    summary=chain2
)

Conditional Routing

from langchain_core.runnables import RunnableBranch

branch = RunnableBranch(
    (lambda x: "simple" in x["query"], simple_chain),
    default_chain
)

Common Questions

Q1: Why use | instead of .pipe()?

• | is more concise and follows Unix pipe intuition
• Clear semantics: data flows left to right

Q2: Does LCEL support async?

• Native support, .ainvoke() / .abatch() automatically use async

Q3: How to debug LCEL chains?

• Use .with_config({"run_name": "StepName"}) to add names
• Insert checkpoints to inspect outputs

References

• LangChain LCEL Documentation
• LangChain GitHub