Quickstart

A LangChain application is usually a composition of small pieces: a prompt template, a chat model, and something that shapes the output. Each piece is a runnable, and runnables compose with the pipe operator.

Your first chain

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain.chat_models import init_chat_model

prompt = ChatPromptTemplate.from_template(
    "Explain {topic} to a {audience} in three sentences."
)
model = init_chat_model("gpt-4o-mini", model_provider="openai")

chain = prompt | model | StrOutputParser()

print(chain.invoke({"topic": "vector search", "audience": "product manager"}))

Streaming the response

Every runnable supports the same interface, so swapping invoke for stream is all it takes to render tokens as they arrive.

for chunk in chain.stream({"topic": "embeddings", "audience": "designer"}):
    print(chunk, end="", flush=True)

What to read next

• Core concepts, for the vocabulary behind runnables and messages.
• Retrieval augmented generation, to ground answers in your own data.
• Agents and tools, when the model needs to decide what to do next.