Skip to content

Latest commit

 

History

History
164 lines (106 loc) · 3.98 KB

agency.md

File metadata and controls

164 lines (106 loc) · 3.98 KB

ReXia.AI Agency and ManagerAgent Classes

Overview

The Agency and ManagerAgent classes are core components of the ReXia.AI framework, responsible for managing a group of autonomous agents to collaboratively execute complex tasks. This documentation provides a detailed overview of these classes, their attributes, methods, and usage.

Agency Class

Agency Class Overview

The Agency class represents a group of autonomous agents capable of collaborating on complex tasks.

Agency Class Attributes

  • main_task: The primary task assigned to the agency.
  • agents: A list of available agents in the agency.
  • manager_llm: The language model used by the manager agent.

Agency Class Methods

Agency Initialization

def __init__(self, main_task: str, agents: List[AgentInfo], manager_llm: RexiaAIOpenAI)

Initializes an Agency instance.

Parameters:

  • main_task: The primary task to be completed by the agency.
  • agents: A list of available agents.
  • manager_llm: The language model used by the manager agent.

Agency Invocation

def invoke(self, task: str = None) -> str

Starts the collaborative task execution, optionally accepting a new task to override the initial one.

Parameters:

  • task: (Optional) A new task to override the initial one.

Returns:

  • str: The results of the task execution.

Agency Usage Example

from rexia_ai.agency import Agency
from rexia_ai.llms import RexiaAIOpenAI

# Initialize the language model
manager_llm = RexiaAIOpenAI(...)

# Create an Agency instance
agency = Agency(
    main_task="Complete the project report",
    agents=[Agent1, Agent2, Agent3],
    manager_llm=manager_llm
)

# Invoke the agency to perform the task
result = agency.invoke()
print(result)

ManagerAgent Class

ManagerAgent Class Overview

The ManagerAgent class manages interactions between agents within an agency.

ManagerAgent Class Attributes

  • agents: A list of available agents.
  • manager_llm: The language model used by the manager.
  • collaboration_channel: A channel for agent collaboration.
  • task: The main task assigned to the manager.
  • subtasks: A list of subtasks derived from the main task.

ManagerAgent Class Methods

ManagerAgent Initialization

def __init__(self, agents: List[AgentInfo], manager_llm: RexiaAIOpenAI)

Initializes the ManagerAgent instance.

Parameters:

  • agents: A list of available agents.
  • manager_llm: The language model used by the manager.

Task Assignment

def assign_task(self, task: str) -> None

Assigns the main task to the manager.

Parameters:

  • task: The main task to be completed.

Agent Management

def manage_agents(self) -> None

Manages the work of various agents to iteratively complete the task.

Results Presentation

def present_results(self) -> str

Summarizes and presents the results of the collaborative task execution.

Returns:

  • str: A summarized report of the collaborative task execution.

Internal Methods

  • _get_next_action() -> Dict[str, Any]: Determines the next action (subtask or completion).
  • _execute_assignment(assignment: AgentAssignment) -> None: Executes a single agent assignment.
  • _create_action_prompt() -> str: Generates prompts for the language model.
  • _summarise_results(results: str) -> str: Summarizes the results of subtask executions.

ManagerAgent Usage Example

from rexia_ai.manager_agent import ManagerAgent
from rexia_ai.llms import RexiaAIOpenAI

# Initialize the language model
manager_llm = RexiaAIOpenAI(...)

# Create a ManagerAgent instance
manager_agent = ManagerAgent(
    agents=[AgentInfo1, AgentInfo2, AgentInfo3],
    manager_llm=manager_llm
)

# Assign a task to the manager
manager_agent.assign_task("Complete the project report")

# Manage agents to perform the task
manager_agent.manage_agents()

# Present the results
results = manager_agent.present_results()
print(results)