agent.py

# agent.py — final version using FunctionCallingAgent

import os
import asyncio
from llama_index.llms.openai import OpenAI
from llama_index.core.agent import FunctionCallingAgent
from llama_index.core.tools import FunctionTool

from langchain_community.utilities.wikipedia import WikipediaAPIWrapper
from langchain_community.document_loaders import YoutubeLoader
from langchain_experimental.tools.python.tool import PythonREPLTool

import whisper
import openpyxl

# Check OpenAI key
if os.getenv("OPENAI_API_KEY"):
    print("✅ Detected OPENAI_API_KEY")
else:
    print("⚠️ Missing OPENAI_API_KEY – LLM may fail")

# Tool 1 — Wikipedia
wiki_api = WikipediaAPIWrapper(top_k_results=1, doc_content_chars_max=1000)

def search_wikipedia(query: str) -> str:
    """Search Wikipedia for a given query and return relevant summary."""
    return wiki_api.run(query)

# Tool 2 — Python with output
python_tool = PythonREPLTool()

def run_python_code(code: str) -> str:
    """Run Python code and return printed result."""
    try:
        if "print(" not in code:
            code = f"print({code})"
        return python_tool.run(code)
    except Exception as e:
        return f"[PYTHON ERROR] {e}"

# Tool 3 — YouTube transcript

def get_youtube_transcript(url: str) -> str:
    """Get transcript from YouTube video."""
    try:
        loader = YoutubeLoader.from_youtube_url(url, add_video_info=False)
        docs = loader.load()
        return " ".join(d.page_content for d in docs)
    except Exception as e:
        return f"[YOUTUBE ERROR] {e}"

# Tool 4 — Whisper transcription

def transcribe_audio(file_path: str) -> str:
    """Transcribe an MP3 file to text using Whisper."""
    try:
        model = whisper.load_model("base")
        res = model.transcribe(file_path)
        return res["text"]
    except Exception as e:
        return f"[AUDIO ERROR] {e}"

# Tool 5 — Excel parser

def extract_excel_total_food_sales(file_path: str) -> str:
    """Sum sales from Excel where category is 'food'."""
    try:
        wb = openpyxl.load_workbook(file_path)
        sheet = wb.active
        total = 0.0
        for _, category, amount in sheet.iter_rows(min_row=2, values_only=True):
            if isinstance(category, str) and "food" in category.lower():
                total += float(amount or 0)
        return f"${total:.2f}"
    except Exception as e:
        return f"[EXCEL ERROR] {e}"

# Assemble tools with return_direct
TOOLS = [
    FunctionTool.from_defaults(search_wikipedia, return_direct=True, name="search_wikipedia"),
    FunctionTool.from_defaults(run_python_code, return_direct=True, name="run_python"),
    FunctionTool.from_defaults(get_youtube_transcript, return_direct=True, name="get_youtube_transcript"),
    FunctionTool.from_defaults(transcribe_audio, return_direct=True, name="transcribe_audio"),
    FunctionTool.from_defaults(extract_excel_total_food_sales, return_direct=True, name="extract_excel_total_food_sales")
]

# Create agent with FunctionCallingAgent
llm = OpenAI(model="gpt-4")

agent = FunctionCallingAgent.from_tools(
    tools=TOOLS,
    llm=llm,
    system_prompt="""
You are a highly capable AI agent taking the GAIA benchmark test.

You have access to the following tools:
- Wikipedia search for factual lookups
- Python runner for math, logic, or text analysis
- YouTube transcript fetcher (via URL)
- Audio transcriber (Whisper, MP3)
- Excel food sales analyzer

Rules:
1. Always try to use a tool if relevant.
2. Return ONLY the final answer in the requested format.
3. Do not guess. If a tool fails, say \"Tool not available\".
4. Follow formats strictly: comma-separated lists, numeric values, chess notation, names only, etc.
5. Avoid all explanation unless requested.
"""
)

# Agent answer wrapper
async def answer_question(question: str) -> str:
    try:
        response = await agent.achat(question)
        return response.response.strip()
    except Exception as e:
        print("❌ Agent error:", e)
        return "[ERROR] " + str(e)