​

marvin.prompts

Prompt system for Marvin.

This module provides a flexible prompt system that supports:

1. Prompts defined as strings or paths to template files
2. Static type checking of prompt variables
3. Dynamic prompt creation from function docstrings and signatures
4. Serialization of prompts and their required attributes

​

Classes

​

Prompt

class Prompt(source: str | Path, _extra_fields: dict[str, Any] = dict())

Base class for prompts.

Prompts can be defined either as strings or paths to template files. Additional attributes can be added by subclassing and will be type-checked.

Methods:

• from_fn

  def from_fn(cls, fn: Callable[..., Any]) -> type[Prompt]
  

  Create a Prompt class from a function’s docstring and signature.

  Args: fn: The function to create a prompt from. The function’s docstring will be used as the template, and its signature will define the required attributes.

  Example: def greet(name: str, age: int): ''' SYSTEM: You are a friendly assistant. USER: Hi , how old are you? ASSISTANT: I’m an AI, but you’re years old! ''' pass

  GreetPrompt = Prompt.from_fn(greet) prompt = GreetPrompt(name=“Alice”, age=30) messages = prompt.to_messages()

• to_messages

  def to_messages(self, **kwargs: Any = {}) -> list[PydanticAIMessage]
  

  Convert the prompt to a list of messages with roles.

  The template can contain role markers (SYSTEM:, USER:, ASSISTANT:) to indicate message roles. Text without a role marker is treated as a user message.

  Example template: ''' SYSTEM: You are a helpful assistant. USER: Hi ! ASSISTANT: Hello ! How can I help you? '''

​

Template

class Template(source: str | Path)

A template for generating prompts.

Args: source: Either a string template or a Path to a template file

Methods:

• render

  def render(self, **kwargs: Any = {}) -> str
  

  Render the template with variables.

Parent Module: marvin