The @supervise decorator is a powerful feature that adds supervision functionality to your functions or tools. It allows you to:

  • Enforce constraints and validations during function execution.
  • Mock function responses for testing and development.
  • Reuse previous execution logs to simulate tool responses.

​
Usage

You can use the @supervise decorator to wrap your functions and specify supervision functions, mock policies, and responses.

from asteroid_sdk.supervision import supervise, MockPolicy

@supervise(
    supervision_functions=[supervisor_function],  # Replace with your supervisor functions
    mock_policy=MockPolicy.NO_MOCK,  # Choose an appropriate mock policy
    mock_responses=None  # Provide mock responses if needed
)
def my_function(args):
    # Function implementation
    pass

​
Example

Here’s an example using the @supervise decorator to supervise a function:

from asteroid_sdk.supervision import supervise, human_supervisor, llm_supervisor

# Define a policy or instructions for the LLM supervisor
EMAIL_POLICY = "Ensure that the email content is appropriate and follows company guidelines."

@supervise(
    supervision_functions=[
        [llm_supervisor(instructions=EMAIL_POLICY)],  # First supervision chain
        [human_supervisor()]  # Second supervision chain
    ]
)
def send_email(to: str, subject: str, body: str):
    """
    Send an email to the specified recipient.

    Parameters:
        to (str): Recipient's email address.
        subject (str): Subject of the email.
        body (str): Body content of the email.
    """
    # Implement the email sending logic
    pass

In this example, the send_email function is supervised by an LLM supervisor, which checks the email content according to the provided policy, and a human supervisor for final approval.

Projects, Tasks & RunsSupervisors Overview