Prompt templates

A prompt template formats named inputs into a prompt value. ChatPromptTemplate preserves message roles instead of building one large string by hand.

Tiny example

from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages([
    ("system", "Answer only from this policy:\n{policy}"),
    ("human", "Order {order_id}: {question}"),
])

messages = prompt.invoke({
    "policy": "Returns are accepted for 30 days.",
    "order_id": "A-19",
    "question": "Can I return it after 45 days?",
})

Remember the flow

1. Write stable instructions once.
2. Name each changing value.
3. Format with a dictionary.
4. Pass the result to a model or compose it with |.

Failure note

Templates format text; they do not make untrusted text safe. Retrieved pages and user input can contain instructions. Keep authorization outside the prompt, mark data clearly, and test prompt-injection cases.

Related

Messages · Runnables and LCEL · Structured output