Prompt Templates

Where we left off

Lesson 2's flow sent whatever you typed straight to Gemini, no shaping at all. Most real flows want to wrap the user's input in something, instructions, examples, formatting, before it reaches the model. That's what the Prompt Template component is for.

Do this yourself

1. Start a new flow (or clear Lesson 2's canvas).
2. Drag a Chat Input onto the canvas.

3. From Models & Agents, drag a Prompt Template component onto the canvas. In its Template field, type:

   Answer as if you were a pirate.
   
   User: {user_input}
   
   Answer:

   The moment you type {user_input}, Langflow adds a new input field to the component named user_input, matching what's in the braces. Compare to this screenshot of exactly this:

   
4. Connect Chat Input's output to the Prompt Template's new user_input input.
5. Add a Google Generative AI component (same setup as Lesson 2: your API key, model set to gemini-3.5-flash-lite) and a Chat Output, and connect Prompt Template's Prompt output into the Google Generative AI component's Input, then its output into Chat Output.
6. Open the Playground, ask something ordinary ("What's the weather like?"), and confirm the answer comes back in pirate voice, that's the template doing its job.

The code, piece by piece

prompt = PromptComponent()
prompt.set(
    template="Answer as if you were a pirate.\n\nUser: {user_input}\n\nAnswer:",
    user_input=chat_input.message_response,
)

template is a plain string with {user_input} inside it, the same {name} syntax as Python's own str.format(). Setting user_input= to chat_input.message_response is the code form of the connection you dragged in the browser: whatever Chat Input produces fills in that slot every time the flow runs.

gemini.set(input_value=prompt.build_prompt)

build_prompt is the Prompt component's output, the fully filled-in string (template with {user_input} replaced), wrapped as a Message so it can feed straight into Gemini's input_value, the same Message-typed contract from Lesson 3.