multiagents_old.py

# a multi agent proposal to solve HF agent course final assignment
import os
import dotenv
from smolagents import CodeAgent
#from smolagents import OpenAIServerModel
from tools.fetch import fetch_webpage, search_web
from smolagents import PythonInterpreterTool, InferenceClientModel
from tools.yttranscript import get_youtube_transcript, get_youtube_title_description
from tools.stt import get_text_transcript_from_audio_file
from tools.image import analyze_image
from common.mylogger import mylog
from smolagents import LiteLLMModel  # Import LiteLLMModel instead of OpenAIServerModel
import os
#from huggingface_hub import InferenceClient
import myprompts

from groq_api import GrokApi


dotenv.load_dotenv()

# gemini_model = OpenAIServerModel(
#     model_id="gemini-2.0-flash",
#     api_key=os.environ["GEMINI_API_KEY"],
#     # Google Gemini OpenAI-compatible API base URL
#     api_base="https://generativelanguage.googleapis.com/v1beta/openai/",
# )

# vllm_model = OpenAIServerModel(
#     model_id="Qwen/Qwen2.5-1.5B-Instruct",
#     api_base="http://192.168.1.39:18000/v1",  
#     api_key="token-abc123",  
# )

# openai_41nano_model = OpenAIServerModel(
#     model_id="gpt-4.1-nano",
#     api_base="https://api.openai.com/v1",
#     api_key=os.environ["OPENAI_API_KEY"],
# )

# openai_41mini_model = OpenAIServerModel(
#     model_id="gpt-4.1-mini",
#     api_base="https://api.openai.com/v1",
#     api_key=os.environ["OPENAI_API_KEY"],
# )



# # --- Agent wrappers ---
# groq_model = LiteLLMModel(
#     model_id="groq/qwen3-32b",  # or any other Groq model like groq/mixtral-8x7b-32768
#     api_key = os.getenv("GROQ_API_KEY"),
#     temperature=0.1,
#     max_tokens=4000,
# )

# SETUP AND TEST



grok_api_key = os.getenv("groq_api")

#InferenceClientModel InferenceClient
My_Agent = InferenceClientModel(
    provider="groq",
    api_key=grok_api_key,
    model_id = "qwen/qwen3-32b"
)

# test_messages = [{"role": "user", "content": "What are the 3 laws of robotics"}]
# output_test = My_Agent(test_messages)
# print('1', output_test)



# My_Agent = client.chat.completions.create(
#     model="qewn/qwen3-32b",
#     messages=[
#         {
#             "role": "user",
#             "content": "How many 'G's in 'huggingface'?"
#         }
#     ],
# )



def check_final_answer(final_answer, agent_memory)  -> bool:
    """
    Check if the final answer is correct.
    basic check on the length of the answer.
    """
    mylog("check_final_answer", final_answer)
    # if return answer is more than 200 characters, we will assume it is not correct    
    if len(str(final_answer)) > 200:
        return False
    else:
        return True


web_agent = CodeAgent(
    model=My_Agent,
    tools=[
        search_web,
        fetch_webpage,                
    ],
    name="web_agent",
    description="Use search engine to find webpages related to a subject and get the page content",
    additional_authorized_imports=["pandas", "numpy","bs4"],
    verbosity_level=1,    
    max_steps=7,
)

audiovideo_agent = CodeAgent(
    model=My_Agent,
    tools=[
        get_youtube_transcript,
        get_youtube_title_description,
        get_text_transcript_from_audio_file,
        analyze_image
    ],
    name="audiovideo_agent",
    description="Extracts information from image, video or audio files from the web",
    additional_authorized_imports=["pandas", "numpy","bs4", "requests"],
    verbosity_level=1,
    max_steps=7,
)



manager_agent = CodeAgent(
    model=My_Agent,
    tools=[ PythonInterpreterTool()],
    managed_agents=[web_agent, audiovideo_agent],    
    additional_authorized_imports=["pandas", "numpy","bs4"],
    planning_interval=5,
    verbosity_level=2,
    final_answer_checks=[check_final_answer],
    max_steps=15,
    name="manager_agent",
    description="A manager agent that coordinates the work of other agents to answer questions.",
)

class MultiAgent:
    def __init__(self):
        print("BasicAgent initialized.")

    def __call__(self, question: str) -> str:
        mylog(self.__class__.__name__, question)        

        try:
            prefix = """You are the top agent of a multi-agent system that can answer questions by coordinating the work of other agents.
            You will receive a question and you will decide which agent to use to answer it.
            You can use the web_agent to search the web for information and for fetching the content of a web page, or the audiovideo_agent to extract information from video or audio files.
            You can also use your own knowledge to answer the question.
            You need to respect the output format that is given to you.
            Finding the correct answer to the question need reasoning and plannig, read the question carrefully, think step by step and do not skip any steps.
            """

            question = prefix + "\nTHE QUESTION:\n" + question + '\n' + myprompts.output_format

            fixed_answer = ""

            fixed_answer = manager_agent.run(question)
            
            return fixed_answer
        except Exception as e:
            error = f"An error occurred while processing the question: {e}"
            print(error)
            return error


if __name__ == "__main__":
    # Example usage

    asyncio.run(main())
    question = """
What was the actual enrollment of the Malko competition in 2023?
"""
    agent = MultiAgent()
    answer = agent(question)
    print(f"Answer: {answer}")