app.py

import os
import gradio as gr
import requests
import pandas as pd
from datetime import datetime
from smolagents import Tool, ToolCallingAgent
from smolagents.models import InferenceClientModel

# Constants
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"

# --- Custom Tools ---
class CalculatorTool(Tool):
    name = "calculator"
    description = "Performs mathematical calculations"
    input_schema = {
        "expression": {
            "type": "string",
            "description": "Math expression to evaluate (e.g. '2+2')"
        }
    }
    output_schema = {
        "result": {
            "type": "string",
            "description": "The result of the expression"
        }
    }

    def use(self, expression: str) -> dict:
        try:
            return {"result": str(eval(expression))}
        except Exception as e:
            return {"result": f"Error: {e}"}


class TimeTool(Tool):
    name = "current_time"
    description = "Gets current UTC time"
    input_schema = {}
    output_schema = {
        "time": {
            "type": "string",
            "description": "Current time in UTC"
        }
    }

    def use(self) -> dict:
        return {"time": datetime.utcnow().strftime("%Y-%m-%d %H:%M:%S UTC")}

# --- Agent ---
class LocalAgent:
    def __init__(self):
        self.tools = [CalculatorTool(), TimeTool()]
        self.agent = ToolCallingAgent(
            tools=self.tools,
            model=InferenceClientModel(
                model_id="HuggingFaceH4/zephyr-7b-beta",
                api_base="https://api-inference.huggingface.co/models"
            )
        )

    def __call__(self, question: str) -> str:
        question_lower = question.lower()
        if any(op in question_lower for op in ["calculate", "+", "-", "*", "/", "what is"]):
            return CalculatorTool().use(question.replace("?", ""))["result"]
        if "time" in question_lower:
            return TimeTool().use()["time"]
        try:
            return str(self.agent.run(question))
        except Exception as e:
            return f"Error: {e}"

# --- Evaluation Function ---
def run_and_submit_all(profile: gr.OAuthProfile | None):
    if not profile:
        return "Please login first.", None

    space_id = os.getenv("SPACE_ID", "local-test")
    api_url = os.getenv("API_URL", DEFAULT_API_URL)

    try:
        agent = LocalAgent()
        questions = requests.get(f"{api_url}/questions", timeout=15).json()

        answers = []
        logs = []

        for q in questions:
            try:
                ans = agent(q["question"])
                answers.append({
                    "task_id": q["task_id"],
                    "submitted_answer": ans
                })
                logs.append({
                    "Task ID": q["task_id"],
                    "Question": q["question"],
                    "Answer": ans
                })
            except Exception as e:
                logs.append({
                    "Task ID": q["task_id"],
                    "Question": q["question"],
                    "Answer": f"Error: {e}"
                })

        submission = {
            "username": profile.username,
            "agent_code": f"https://huggingface.co/spaces/{space_id}",
            "answers": answers
        }

        result = requests.post(f"{api_url}/submit", json=submission, timeout=60).json()

        return (
            f"✅ Score: {result.get('score', 'N/A')}%\n"
            f"Correct: {result.get('correct_count', '?')}/{result.get('total_attempted', '?')}",
            pd.DataFrame(logs)
        )

    except Exception as e:
        return f"Evaluation failed: {e}", pd.DataFrame([])

# --- Gradio UI ---
with gr.Blocks(title="Agent Evaluation") as app:
    gr.Markdown("## 🤖 Agent Evaluation with smolagents")
    gr.Markdown("Login, then click 'Run Evaluation' to test your agent.")

    gr.LoginButton()               # Login button visible
    profile = gr.OAuthProfile()   # Profile input (not visible)

    run_btn = gr.Button("🚀 Run Evaluation")
    output = gr.Textbox(label="Evaluation Result")
    results_table = gr.DataFrame(label="Answer Log")

    run_btn.click(
        fn=run_and_submit_all,
        inputs=[profile],          # ✅ Must include this
        outputs=[output, results_table]
    )

if __name__ == "__main__":
    app.launch()