How to create a prompt template in LangChain

A chat model receives ordered messages rather than one undifferentiated block of text. A reusable LangChain prompt template preserves those message roles while leaving selected details open for each invocation.

The ChatPromptTemplate class represents the system and human messages separately, and placeholders such as {change} become required input variables. It is available from langchain_core.prompts without a model-provider package.

Formatting the template locally does not call a model or require an API key. A successful invocation returns a ChatPromptValue whose messages show the exact text that a later model call would receive; a missing required value raises an error before any provider request is made.

1. Create prompt_template.py with the message roles and placeholders for the reusable prompt.

   $ cat > prompt_template.py <<'PY'
   from langchain_core.prompts import ChatPromptTemplate
    
   prompt = ChatPromptTemplate.from_messages(
       [
           ("system", "You write release notes for infrastructure teams."),
           (
               "human",
               "Summarize {change} for {audience} in no more than {sentences} sentences.",
           ),
       ]
   )
   PY

   The default f-string template format substitutes named values without evaluating a template language. Jinja2 templates from untrusted sources can execute unsafe expressions and should not be accepted as user input.

2. Append the runtime values and template invocation to prompt_template.py.

   $ cat >> prompt_template.py <<'PY'
    
   values = {
       "change": "automated snapshot cleanup",
       "audience": "database operators",
       "sentences": 2,
   }
   prompt_value = prompt.invoke(values)
   PY

3. Append the output loop that prints each formatted message with its role.

   $ cat >> prompt_template.py <<'PY'
    
   print("Required variables:", ", ".join(prompt.input_variables))
   for message in prompt_value.to_messages():
       print(f"{message.type}: {message.content}")
   PY

4. Confirm prompt_template.py contains the completed template, values, invocation, and output loop.

   prompt_template.py

   from langchain_core.prompts import ChatPromptTemplate
    
   prompt = ChatPromptTemplate.from_messages(
       [
           ("system", "You write release notes for infrastructure teams."),
           (
               "human",
               "Summarize {change} for {audience} in no more than {sentences} sentences.",
           ),
       ]
   )
    
   values = {
       "change": "automated snapshot cleanup",
       "audience": "database operators",
       "sentences": 2,
   }
   prompt_value = prompt.invoke(values)
    
   print("Required variables:", ", ".join(prompt.input_variables))
   for message in prompt_value.to_messages():
       print(f"{message.type}: {message.content}")

5. Run prompt_template.py to render the fully populated system and human messages.

   $ python3 prompt_template.py
   Required variables: audience, change, sentences
   system: You write release notes for infrastructure teams.
   human: Summarize automated snapshot cleanup for database operators in no more than 2 sentences.