AgentConfig

AgentConfig class is a configuration base class used for setting up agent instances in automata. It is an abstract class (as it is derived from the ABC class) that outlines the structure and basic functionality for agent configurations in the package. This class provides a basis for creating and loading custom agent configurations.

Overview

The AgentConfig class includes several attributes, such as config_name, tools, instructions, description, model, stream, verbose, max_iterations, temperature, session_id. These attributes hold various information needed to configure and run an agent.

Additionally, it has methods that must be implemented in subclasses - setup(), load(), and get_llm_provider(). These methods are crucial for setting up the agent and loading the needed configuration.

A nested Config class is also included, allowing for further configuration settings. Using arbitrary types and specifying the provider for low-level machine (LLM) operations are available within this nested class.

Attributes

config_name: Indicates the name of the agent configuration (default is AgentConfigName.DEFAULT).
tools: A list of Tool instances used by the agent.
instructions: A string containing instructions for the agent.
description: A descriptive string for the agent.
model: The model used by the agent.
stream: Boolean indicating whether streaming functionality is enabled.
verbose: Boolean controlling the verbosity of the agent.
max_iterations: The maximum number of iterations the agent will attempt.
temperature: A float controlling the random behaviour of the agent.
session_id: An optional string describing the current session.

Example Usage

The following example shows how to create a subclass of AgentConfig. In the subclass, abstract methods are implemented and additional configurations are set.

from automata.config.base import AgentConfig, ConfigCategory
from automata.config.base import AgentConfigName, LLMProvider

class CustomAgentConfig(AgentConfig):
    def __init__(self, **data: Any):
        super().__init__(**data)
        self.custom_attribute = ''

    def setup(self) -> None:
        pass

    @classmethod
    def load(cls, config_name: AgentConfigName) -> "CustomAgentConfig":
        loaded_yaml = cls._load_automata_yaml_config(config_name)
        return CustomAgentConfig(**loaded_yaml)

    @staticmethod
    def get_llm_provider() -> LLMProvider:
        return LLMProvider.OPENAI

custom_config = CustomAgentConfig.load(AgentConfigName.AUTOMATA_MAIN)

Limitations

As AgentConfig is an abstract base class, it can’t be instantiated directly. All abstract methods setup(), load(), and get_llm_provider() must be implemented in any subclass. It also assumes a specific file structure for loading YAML configurations and will raise errors if it can’t locate the necessary files.

Follow-up Questions:

How can we handle custom YAML file structures in subclasses of AgentConfig?
Can we add a method to validate the content of the loaded YAML configuration file?