Skip to main content
Version: 1.0

aixplain.modules.team_agent.inspector

Pre-defined agent for inspecting the data flow within a team agent.

WARNING: This feature is currently in private beta.

Example usage:

inspector = Inspector( name="my_inspector", model_id="my_model", model_config={"prompt": "Check if the data is safe to use."}, policy=InspectorPolicy.ADAPTIVE )

team = TeamAgent( name="team" agents=agents, description="team description", llm_id="xyz", use_mentalist=True, inspectors=[inspector], )

AUTO_DEFAULT_MODEL_ID

GPT-4.1 Nano

InspectorAuto Objects

class InspectorAuto(str, Enum)

[view_source]

A list of keywords for inspectors configured automatically in the backend.

get_name

def get_name() -> Text

[view_source]

Get the standardized name for this inspector type.

This method generates a consistent name for the inspector by prefixing the enum value with "inspector_".

Returns:

  • Text - The inspector name in the format "inspector_<type>".

InspectorPolicy Objects

class InspectorPolicy(str, Enum)

[view_source]

Which action to take if the inspector gives negative feedback.

WARN

log only, continue execution

ABORT

stop execution

ADAPTIVE

adjust execution according to feedback

Inspector Objects

class Inspector(ModelWithParams)

[view_source]

Pre-defined agent for inspecting the data flow within a team agent.

The model should be onboarded before using it as an inspector.

Attributes:

  • name - The name of the inspector.
  • model_id - The ID of the model to wrap.
  • model_params - The configuration for the model.
  • policy - The policy for the inspector. Default is ADAPTIVE.

__init__

def __init__(*args, **kwargs)

[view_source]

Initialize an Inspector instance.

This method initializes an inspector with either a custom model or an automatic configuration. If auto is specified, it uses the default auto model ID.

Arguments:

  • *args - Variable length argument list passed to parent class.
  • **kwargs - Arbitrary keyword arguments. Supported keys:
    • name (Text): The inspector's name
    • model_id (Text): The model ID to use
    • model_params (Dict, optional): Model configuration
    • auto (InspectorAuto, optional): Auto configuration type
    • policy (InspectorPolicy, optional): Inspector policy

Notes:

If auto is specified in kwargs, model_id is automatically set to AUTO_DEFAULT_MODEL_ID.

validate_name

@field_validator("name")
def validate_name(cls, v: Text) -> Text

[view_source]

Validate the inspector name field.

This validator ensures that the inspector's name is not empty.

Arguments:

  • v Text - The name value to validate.

Returns:

  • Text - The validated name value.

Raises:

  • ValueError - If the name is an empty string.