main.py

import autogen
import dotenv

from src.autopilot import AutoPilotAgent

dotenv.load_dotenv()

config_list = autogen.config_list_from_dotenv(
    ".env",
    {"gpt-4o": "OPENAI_API_KEY"}
)

llm_config = {
    "cache_seed": 43,
    "temperature": 0,
    "config_list": config_list,
    "timeout": 120,
}

user_proxy = autogen.UserProxyAgent(
    name="Admin",
    system_message="A human admin.",
    code_execution_config={
        "work_dir": "paper",
        "use_docker": False
    },
    human_input_mode="TERMINATE",
)

engineer = autogen.AssistantAgent(
    name="Engineer",
    llm_config=llm_config,
    system_message="""Engineer. You follow an approved plan. Make sure you save code to disk.  
    You write python/shell code to solve tasks. Wrap the code in a code block that specifies
    the script type and the name of the file to save to disk.""",
)

scientist = autogen.AssistantAgent(
    name="Scientist",
    llm_config=llm_config,
    system_message="""Scientist. You follow an approved plan. But first ask the Admin's expectations! 
    You are able to categorize papers after seeing their abstracts printed. You don't write code.""",
)

planner = autogen.AssistantAgent(
    name="Planner",
    system_message="""Planner. Suggest a plan. Revise the plan based on feedback from admin and critic, 
    until admin approval. The plan may involve an engineer who can write code and a scientist who doesn't write code. 
    Explain the plan first. Be clear which step is performed by an engineer, and which step is performed by a scientist.
""",
    llm_config=llm_config,
)

critic = autogen.AssistantAgent(
    name="Critic",
    system_message="Critic. Double check plan, claims, code from other agents and provide feedback. "
                   "Check whether the plan includes adding verifiable info such as source URL.",
    llm_config=llm_config,
)

autopilot = AutoPilotAgent(
    name="AutoPilot",
    llm_config=llm_config,
)

group_chat = autogen.GroupChat(
    agents=[user_proxy, planner, scientist, engineer, critic, autopilot],
    messages=[],
    max_round=7,
    speaker_selection_method='round_robin',
)

manager = autogen.GroupChatManager(groupchat=group_chat, llm_config=llm_config)

prompt = """
Could you please write me an article about Agentic AI?
"""

similar_found = autopilot.reapply_steps_if_found(prompt, threshold=0.7)

if not similar_found:
    user_proxy.initiate_chat(manager, message=prompt)
    autopilot.store_prompt_and_last_message_in_db(prompt)

autopilot.close_db()