Pure Python#

When to use#

The task is pure deterministic logic with no need for LLM reasoning. For example:

• Word counting
• File reading / writing
• Data format conversion
• Math

Design points#

• Do not use the Agent method execution
• Do not call llm()
• No runtime parameter needed
• Use a standard Google-style docstring

Examples#

def word_count(text: str) -> int:
    """Count the number of words in a text.

    Args:
        text: Input text.

    Returns:
        The word count.
    """
    return len(text.split())

def extract_emails(text: str) -> list[str]:
    """Extract every email address from a text.

    Args:
        text: Input text.

    Returns:
        List of email addresses.
    """
    import re
    return re.findall(r'[\w.+-]+@[\w-]+\.[\w.-]+', text)

Session DAG#

Ordinary helpers in managed Program sources and ordinary Agent methods receive call scopes automatically. Arbitrary host functions do not receive source capture. Deterministic behavior alone does not determine whether a call appears in the DAG.

Use an ordinary Agent method when a deterministic helper belongs to a recorded workflow. It does not need a model request or tool registration. Set tool=True only when the model should be allowed to call it.

from openprogram import Agent

class TextAgent(Agent):
    def word_count(self, text: str) -> int:
        """Count words."""
        return len(text.split())

Deterministic and model operations#