How to render prompt templates with PromptBuilder in Haystack

Language-model components accept a finished prompt string, while application data usually arrives as documents, questions, and other separate values. PromptBuilder converts those values into one string by rendering a Jinja template before the prompt reaches a generator.

Running PromptBuilder by itself exposes the rendered text without requiring an API key or model request. The same prompt output can later connect to a generator inside a Haystack pipeline.

Every value referenced by the template is marked as required. Setting required_variables to * makes a missing documents or question input stop the run instead of silently rendering an incomplete prompt.

1. Create prompt_builder_demo.py with the Haystack imports and runtime values.

   prompt_builder_demo.py

   from haystack import Document
   from haystack.components.builders import PromptBuilder
    
    
   documents = [
       Document(content="Refund requests require a matching billing record."),
       Document(content="Account ownership changes require manager approval."),
   ]
   question = "What must support check before approving a refund?"

2. Append the Jinja prompt template after the question assignment.

   jinja_open = "{" + "{"
   jinja_close = "}" + "}"
    
   template = (
       "Answer the question using only the context.\n\n"
       "Context:\n"
       "{% for document in documents %}\n"
       "- " + jinja_open + " document.content " + jinja_close + "\n"
       "{% endfor %}\n\n"
       "Question: " + jinja_open + " question " + jinja_close + "\n"
       "Answer:"
   )

   The loop reads each Document.content value, while the question expression inserts the question once after the context.

3. Append the rendering section after the template.

   builder = PromptBuilder(template=template, required_variables="*")
   result = builder.run(documents=documents, question=question)
    
   print(result["prompt"])

   PromptBuilder.run() returns a dictionary whose prompt value contains the rendered string expected by a text generator.

4. Run prompt_builder_demo.py to verify that both documents and the question replace their template variables.

   $ python3 prompt_builder_demo.py
   Answer the question using only the context.
    
   Context:
    
   - Refund requests require a matching billing record.
    
   - Account ownership changes require manager approval.
    
    
   Question: What must support check before approving a refund?
   Answer: