import os
import gradio as gr
import requests
import inspect
import pandas as pd
import asyncio
import aiohttp
import time
import random
import json
import google.generativeai as genai
import tempfile
import uuid
import math
import cmath
import numpy as np
from urllib.parse import urlparse
from smolagents import FinalAnswerTool, Tool, tool, OpenAIServerModel, DuckDuckGoSearchTool, CodeAgent, VisitWebpageTool
from dotenv import load_dotenv
load_dotenv()
# (Keep Constants as is)
# --- Constants ---
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
OPENAI_TOKEN = os.getenv("OPENAI_API_KEY")
GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
GOOGLE_SEARCH_API_KEY = os.getenv('GOOGLE_SEARCH_API_KEY')
GOOGLE_SEARCH_CX = os.getenv('GOOGLE_SEARCH_CX')
# Configure Gemini
if GOOGLE_API_KEY:
genai.configure(api_key=GOOGLE_API_KEY)
# --- Custom Tools ---
class GoogleSearchTool(Tool):
name = "google_search"
description = "Search Google for current information and facts"
inputs = {"query": {"type": "string", "description": "The search query for Google"}}
output_type = "string"
def __init__(self):
super().__init__()
self.is_initialized = True
self.google_search_api_key = GOOGLE_SEARCH_API_KEY
self.google_search_cx = GOOGLE_SEARCH_CX
def forward(self, query: str) -> str:
"""Perform a Google search using the Custom Search API"""
if not self.google_search_api_key or not self.google_search_cx:
return f"Google Search API not configured. Query was: {query}"
try:
url = "https://www.googleapis.com/customsearch/v1"
params = {
'key': self.google_search_api_key,
'cx': self.google_search_cx,
'q': query,
'num': 5
}
response = requests.get(url, params=params, timeout=10)
if response.status_code != 200:
return f"Google Search failed with status {response.status_code}"
results = response.json()
if 'items' not in results:
return f"No search results found for: {query}"
# Format search results
formatted_results = f"Google search results for '{query}':\n\n"
for item in results['items']:
title = item.get('title', 'No title')
snippet = item.get('snippet', 'No description')
formatted_results += f"• {title}: {snippet}\n"
return formatted_results[:1000] # Limit length
except Exception as e:
return f"Google Search error for '{query}': {str(e)}"
class KnowledgeBaseTool(Tool):
name = "knowledge_base"
description = "Access structured knowledge for common topics"
inputs = {"topic": {"type": "string", "description": "The topic to look up"}}
output_type = "string"
def __init__(self):
super().__init__()
self.is_initialized = True
# Common knowledge base
self.knowledge = {
"olympics": "Olympic Games data: Countries, athletes, years, sports",
"countries": "Country codes: ISO, IOC, FIFA codes and country information",
"sports": "Sports history, rules, famous athletes and events",
"science": "Scientific facts, formulas, discoveries, and researchers",
"history": "Historical events, dates, people, and places",
"geography": "Countries, capitals, populations, and geographical features"
}
def forward(self, topic: str) -> str:
topic_lower = topic.lower()
for key, info in self.knowledge.items():
if key in topic_lower:
return f"Knowledge base: {info}. Use this context to answer questions about {topic}."
return f"No specific knowledge base entry for '{topic}'. Use general reasoning."
class WikipediaSearchTool(Tool):
name = "wikipedia_search"
description = "Search Wikipedia for information"
inputs = {"query": {"type": "string", "description": "The search query for Wikipedia"}}
output_type = "string"
def __init__(self):
super().__init__()
self.is_initialized = True
def forward(self, query: str) -> str:
"""Search Wikipedia with simple fallback."""
try:
import requests
wiki_url = "https://en.wikipedia.org/api/rest_v1/page/summary/" + query.replace(" ", "_")
response = requests.get(wiki_url, timeout=2)
if response.status_code == 200:
data = response.json()
if 'extract' in data and data['extract']:
return f"Wikipedia: {data['extract'][:500]}" # Limit length
except Exception as e:
print(f"Wikipedia search failed: {e}")
return f"Wikipedia search unavailable for '{query}'. Use your knowledge to answer."
# --- Mathematical Tools ---
class MathTool(Tool):
name = "math_calculator"
description = "Perform mathematical calculations including basic operations, powers, roots"
inputs = {"expression": {"type": "string", "description": "Mathematical expression to evaluate"}}
output_type = "string"
def __init__(self):
super().__init__()
self.is_initialized = True
def forward(self, expression: str) -> str:
try:
# Safe evaluation of mathematical expressions
allowed_names = {
k: v for k, v in math.__dict__.items() if not k.startswith("__")
}
allowed_names.update({"abs": abs, "round": round, "pow": pow})
result = eval(expression, {"__builtins__": {}}, allowed_names)
return f"Result: {result}"
except Exception as e:
return f"Math calculation error: {str(e)}"
# --- File Processing Tools ---
class FileProcessorTool(Tool):
name = "file_processor"
description = "Download files from URLs, save content to files, and analyze CSV/Excel files"
inputs = {
"action": {"type": "string", "description": "Action: 'download', 'save', 'analyze_csv', 'analyze_excel'"},
"data": {"type": "string", "description": "URL, content, or file path depending on action"}
}
output_type = "string"
def __init__(self):
super().__init__()
self.is_initialized = True
def forward(self, action: str, data: str) -> str:
try:
if action == "download":
return self._download_file(data)
elif action == "save":
return self._save_content(data)
elif action == "analyze_csv":
return self._analyze_csv(data)
elif action == "analyze_excel":
return self._analyze_excel(data)
else:
return f"Unknown action: {action}"
except Exception as e:
return f"File processing error: {str(e)}"
def _download_file(self, url: str) -> str:
try:
response = requests.get(url, timeout=10)
response.raise_for_status()
filename = os.path.basename(urlparse(url).path) or f"download_{uuid.uuid4().hex[:8]}"
filepath = os.path.join(tempfile.gettempdir(), filename)
with open(filepath, 'wb') as f:
f.write(response.content)
return f"File downloaded to: {filepath}"
except Exception as e:
return f"Download failed: {str(e)}"
def _save_content(self, content: str) -> str:
try:
filepath = os.path.join(tempfile.gettempdir(), f"content_{uuid.uuid4().hex[:8]}.txt")
with open(filepath, 'w') as f:
f.write(content)
return f"Content saved to: {filepath}"
except Exception as e:
return f"Save failed: {str(e)}"
def _analyze_csv(self, filepath: str) -> str:
try:
df = pd.read_csv(filepath)
result = f"CSV Analysis:\n- Rows: {len(df)}\n- Columns: {len(df.columns)}\n"
result += f"- Column names: {', '.join(df.columns)}\n"
result += f"- Summary:\n{df.describe().to_string()}"
return result[:1000]
except Exception as e:
return f"CSV analysis failed: {str(e)}"
def _analyze_excel(self, filepath: str) -> str:
try:
df = pd.read_excel(filepath)
result = f"Excel Analysis:\n- Rows: {len(df)}\n- Columns: {len(df.columns)}\n"
result += f"- Column names: {', '.join(df.columns)}\n"
result += f"- Summary:\n{df.describe().to_string()}"
return result[:1000]
except Exception as e:
return f"Excel analysis failed: {str(e)}"
# --- Code Execution Tool ---
class CodeExecutorTool(Tool):
name = "code_executor"
description = "Execute Python code safely and return results"
inputs = {"code": {"type": "string", "description": "Python code to execute"}}
output_type = "string"
def __init__(self):
super().__init__()
self.is_initialized = True
def forward(self, code: str) -> str:
try:
# Create a restricted execution environment
allowed_modules = {
'math': math, 'cmath': cmath, 'random': random,
'json': json, 'time': time, 'os': os,
'pandas': pd, 'numpy': np
}
# Capture output
import io
import sys
from contextlib import redirect_stdout, redirect_stderr
stdout_capture = io.StringIO()
stderr_capture = io.StringIO()
with redirect_stdout(stdout_capture), redirect_stderr(stderr_capture):
exec(code, {"__builtins__": {}, **allowed_modules})
stdout_result = stdout_capture.getvalue()
stderr_result = stderr_capture.getvalue()
result = "Code executed successfully\n"
if stdout_result:
result += f"Output: {stdout_result}\n"
if stderr_result:
result += f"Errors: {stderr_result}\n"
return result[:1000]
except Exception as e:
return f"Code execution error: {str(e)}"
# --- Web Scraping Tool ---
class WebScrapeTool(Tool):
name = "web_scraper"
description = "Scrape content from web pages"
inputs = {"url": {"type": "string", "description": "URL to scrape"}}
output_type = "string"
def __init__(self):
super().__init__()
self.is_initialized = True
def forward(self, url: str) -> str:
try:
response = requests.get(url, timeout=10, headers={'User-Agent': 'Mozilla/5.0'})
response.raise_for_status()
# Simple text extraction (you could add BeautifulSoup for better parsing)
content = response.text
# Extract title if possible
title_start = content.find('')
title_end = content.find('')
title = "Unknown"
if title_start != -1 and title_end != -1:
title = content[title_start+7:title_end]
# Return basic info
result = f"Page Title: {title}\n"
result += f"Content Length: {len(content)} characters\n"
result += f"First 500 chars: {content[:500]}..."
return result
except Exception as e:
return f"Web scraping error: {str(e)}"
# --- Gemini Model Wrapper ---
class GeminiModel:
def __init__(self, model_name="gemini-2.0-flash", temperature=0.0, max_tokens=500):
self.model_name = model_name
self.temperature = temperature
self.max_tokens = max_tokens
if GOOGLE_API_KEY:
self.model = genai.GenerativeModel(model_name)
else:
self.model = None
print("Warning: Google API key not found, falling back to OpenAI")
def generate_content(self, prompt):
if self.model:
try:
response = self.model.generate_content(
prompt,
generation_config=genai.types.GenerationConfig(
max_output_tokens=self.max_tokens,
temperature=self.temperature
)
)
return response.text
except Exception as e:
print(f"Gemini API error: {e}")
return f"Error generating response: {e}"
else:
return "Gemini model not available"
# --- Basic Agent Definition ---
# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
class SlpMultiAgent:
def __init__(self):
print("BasicAgent initialized with Gemini and Google Search.")
async def __call__(self, question: str) -> str:
print(f"Agent received question (first 50 chars): {question[:50]}...")
fixed_answer = "This is a default answer."
print(f"Agent returning fixed answer: {fixed_answer}")
# Truncate question to avoid exceeding model context length
MAX_QUESTION_LENGTH = 1000
short_question = question # [:MAX_QUESTION_LENGTH]
# Use Gemini as primary model, fallback to OpenAI
if GOOGLE_API_KEY:
model = GeminiModel(
model_name="gemini-2.0-flash",
temperature=0.0,
max_tokens=400
)
print("Using Gemini model")
else:
model = OpenAIServerModel(
model_id="gpt-3.5-turbo",
temperature=0.0,
max_tokens=400
)
print("Using OpenAI model (Gemini not available)")
# Simplified research agent
research_agent = CodeAgent(
tools=[GoogleSearchTool(), KnowledgeBaseTool(),WikipediaSearchTool(),WebScrapeTool()],
model=model if not isinstance(model, GeminiModel) else OpenAIServerModel(model_id="gpt-3.5-turbo", temperature=0.0, max_tokens=400),
additional_authorized_imports=["re", "json"],
max_steps=4,
name="ResearchAgent",
verbosity_level=0,
description="Simple research with Google Search."
)
solver_agent = CodeAgent(
tools=[MathTool(), GoogleSearchTool()],
model=model if not isinstance(model, GeminiModel) else OpenAIServerModel(model_id="gpt-3.5-turbo", temperature=0.0, max_tokens=400),
additional_authorized_imports=["math", "re"],
max_steps=4,
name="SolverAgent",
verbosity_level=0,
description="Simple problem solving with math and search."
)
manager_agent = CodeAgent(
model=model if not isinstance(model, GeminiModel) else OpenAIServerModel(model_id="gpt-3.5-turbo", temperature=0.0, max_tokens=400),
tools=[GoogleSearchTool(), MathTool(), FileProcessorTool(), KnowledgeBaseTool()],
managed_agents=[research_agent, solver_agent],
name="ManagerAgent",
description="Manager with comprehensive tool access and agent coordination.",
additional_authorized_imports=["re", "math", "json", "pandas", "numpy", "requests", "time", "os", "tempfile", "uuid"],
planning_interval=1,
verbosity_level=0,
max_steps=6,
final_answer_checks=[check_reasoning]
)
# Create a task for the agent run with retry mechanism for rate limits
max_retries = 3
result = None
for attempt in range(max_retries):
try:
loop = asyncio.get_event_loop()
result = await loop.run_in_executor(
None,
lambda: manager_agent.run(f"""
Question: {short_question}
CRITICAL INSTRUCTIONS:
1. Use tools DIRECTLY, not in code execution
2. For factual questions: Use google_search() tool immediately
3. For math: Use math_calculator() tool directly
4. DO NOT write complex code - use tools instead
5. DO NOT call ResearchAgent() or SolverAgent() in code
Available tools:
- google_search(query): Search the web
- math_calculator(expression): Calculate math
- file_processor(action, data): Handle files
- knowledge_base(topic): Get knowledge
SIMPLE APPROACH:
- Call the appropriate tool
- Get the result
- Provide final_answer()
Example for factual question:
1. Call google_search("your query")
2. Extract answer from results
3. Call final_answer("the answer")
NO complex code execution. Use tools directly.
ALWAYS end with:
final_answer("your answer here")
""")
)
break # Success, exit retry loop
except Exception as e:
print(f"Attempt {attempt+1}/{max_retries} failed: {e}")
if "rate limit" in str(e).lower() and attempt < max_retries - 1:
# Add jitter to avoid synchronized retries
wait_time = (attempt + 1) * 10 + random.uniform(0, 5)
print(f"Rate limit hit. Waiting {wait_time:.2f} seconds before retry...")
await asyncio.sleep(wait_time)
elif attempt < max_retries - 1:
await asyncio.sleep(5) # Wait before general retry
else:
print(f"All attempts failed. Returning default answer.")
return "I apologize, but I'm currently experiencing technical difficulties. Please try again later."
# If we couldn't get a result after all retries
if result is None:
return "I apologize, but I'm currently experiencing technical difficulties. Please try again later."
# Extract clean answer from result
if result and isinstance(result, str):
# Look for final_answer pattern
import re
final_answer_match = re.search(r'final_answer\(["\']([^"\']*)["\'\)]', result) # Fixed regex
if final_answer_match:
clean_answer = final_answer_match.group(1)
return clean_answer
# If no final_answer found, try to extract the last meaningful line
lines = result.strip().split('\n')
for line in reversed(lines):
line = line.strip()
if line and not line.startswith('#') and not line.startswith('###') and len(line) < 200:
return line
# Return the result from the agent
return result if result else "Unable to determine answer."
def check_reasoning(final_answer, agent_memory):
# Skip expensive validation to save costs
return True
async def run_and_submit_all(profile):
"""
Fetches all questions, runs the BasicAgent on them, submits all answers,
and displays the results asynchronously.
"""
# --- Determine HF Space Runtime URL and Repo URL ---
space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
# Handle different profile types
if profile:
if hasattr(profile, 'username'):
# It's an OAuthProfile object
username = profile.username
else:
# It's a string or other type
username = str(profile)
print(f"User logged in: {username}")
else:
print("User not logged in.")
return "Please Login to Hugging Face with the button.", None
api_url = DEFAULT_API_URL
questions_url = f"{api_url}/questions"
submit_url = f"{api_url}/submit"
# 1. Instantiate Agent ( modify this part to create your agent)
try:
agent = SlpMultiAgent()
except Exception as e:
print(f"Error instantiating agent: {e}")
return f"Error initializing agent: {e}", None
# In the case of an app running as a hugging Face space, this link points toward your codebase ( usefull for others so please keep it public)
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
print(agent_code)
# 2. Fetch Questions
print(f"Fetching questions from: {questions_url}")
try:
async with aiohttp.ClientSession() as session:
async with session.get(questions_url, timeout=15) as response:
response.raise_for_status()
questions_data = await response.json()
if not questions_data:
print("Fetched questions list is empty.")
return "Fetched questions list is empty or invalid format.", None
print(f"Fetched {len(questions_data)} questions.")
except aiohttp.ClientError as e:
print(f"Error fetching questions: {e}")
return f"Error fetching questions: {e}", None
except ValueError as e: # JSON decode error
print(f"Error decoding JSON response from questions endpoint: {e}")
return f"Error decoding server response for questions: {e}", None
except Exception as e:
print(f"An unexpected error occurred fetching questions: {e}")
return f"An unexpected error occurred fetching questions: {e}", None
# 3. Run your Agent
results_log = []
answers_payload = []
print(f"Running agent on {len(questions_data)} questions...")
# Process questions one at a time to avoid rate limits
semaphore = asyncio.Semaphore(1) # Process 1 question at a time
async def process_question(item):
task_id = item.get("task_id")
question_text = item.get("question")
if not task_id or question_text is None:
print(f"Skipping item with missing task_id or question: {item}")
return None
async with semaphore:
max_retries = 3
for attempt in range(max_retries):
try:
print(f"Processing task {task_id}, attempt {attempt+1}/{max_retries}")
submitted_answer = await agent(question_text)
return {"task_id": task_id, "submitted_answer": submitted_answer,
"log": {"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer}}
except Exception as e:
print(f"Error running agent on task {task_id}, attempt {attempt+1}: {e}")
if "rate limit" in str(e).lower() and attempt < max_retries - 1:
# Exponential backoff with jitter
wait_time = (2 ** attempt) * 5 + random.uniform(0, 3)
print(f"Rate limit hit. Waiting {wait_time:.2f} seconds before retry...")
await asyncio.sleep(wait_time)
elif attempt < max_retries - 1:
await asyncio.sleep(5) # Reduced wait time
else:
# All retries failed, return default answer
default_answer = "This is a default answer."
return {"task_id": task_id, "submitted_answer": default_answer,
"log": {"Task ID": task_id, "Question": question_text, "Submitted Answer": default_answer}}
# Create tasks for all questions
tasks = [process_question(item) for item in questions_data]
results = await asyncio.gather(*tasks)
# Process results
for result in results:
if result is not None:
answers_payload.append({"task_id": result["task_id"], "submitted_answer": result["submitted_answer"]})
results_log.append(result["log"])
if not answers_payload:
print("Agent did not produce any answers to submit.")
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
# 4. Prepare Submission
submission_data = {"username": str(username).strip(), "agent_code": agent_code, "answers": answers_payload}
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
print(status_update)
# 5. Submit
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
try:
async with aiohttp.ClientSession() as session:
async with session.post(submit_url, json=submission_data, timeout=60) as response:
response.raise_for_status()
result_data = await response.json()
final_status = (
f"Submission Successful!\n"
f"User: {result_data.get('username')}\n"
f"Overall Score: {result_data.get('score', 'N/A')}% "
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
f"Message: {result_data.get('message', 'No message received.')}"
)
print("Submission successful.")
results_df = pd.DataFrame(results_log)
return final_status, results_df
except aiohttp.ClientResponseError as e:
error_detail = f"Server responded with status {e.status}."
try:
error_text = await e.response.text()
try:
error_json = await e.response.json()
error_detail += f" Detail: {error_json.get('detail', error_text)}"
except ValueError:
error_detail += f" Response: {error_text[:500]}"
except:
pass
status_message = f"Submission Failed: {error_detail}"
print(status_message)
results_df = pd.DataFrame(results_log)
return status_message, results_df
except asyncio.TimeoutError:
status_message = "Submission Failed: The request timed out."
print(status_message)
results_df = pd.DataFrame(results_log)
return status_message, results_df
except aiohttp.ClientError as e:
status_message = f"Submission Failed: Network error - {e}"
print(status_message)
results_df = pd.DataFrame(results_log)
return status_message, results_df
except Exception as e:
status_message = f"An unexpected error occurred during submission: {e}"
print(status_message)
results_df = pd.DataFrame(results_log)
return status_message, results_df
# --- Build Gradio Interface using Blocks ---
with gr.Blocks() as demo:
gr.Markdown("# Enhanced Agent with Google Search & Gemini")
gr.Markdown(
"""
**Features:**
- **Google Search Integration**: Primary tool for factual information
- **Gemini 2.0 Flash**: Advanced AI model for reasoning
- **Multi-Agent Architecture**: Research and Solver agents with search capabilities
**Instructions:**
1. Set up your environment variables: GOOGLE_API_KEY, GOOGLE_SEARCH_API_KEY, GOOGLE_SEARCH_CX
2. Log in to your Hugging Face account using the button below
3. Click 'Run Evaluation & Submit All Answers' to start the enhanced agent
---
**Note:** The agent will prioritize Google Search for factual questions, providing more accurate and current information.
"""
)
login_button = gr.LoginButton()
run_button = gr.Button("Run Evaluation & Submit All Answers")
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
# Removed max_rows=10 from DataFrame constructor
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
def sync_wrapper(profile):
# This wrapper ensures we have access to the profile
if not profile:
print("No profile available in sync_wrapper")
return "Please Login to Hugging Face with the button.", None
print(f"Profile type in wrapper: {type(profile)}")
try:
return asyncio.run(run_and_submit_all(profile))
except Exception as e:
print(f"Error in sync_wrapper: {e}")
return f"Error processing request: {e}", None
run_button.click(
fn=sync_wrapper,
inputs=login_button,
outputs=[status_output, results_table]
)
if __name__ == "__main__":
print("\n" + "-"*30 + " App Starting " + "-"*30)
# Check for SPACE_HOST and SPACE_ID at startup for information
space_host_startup = os.getenv("SPACE_HOST")
space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
if space_host_startup:
print(f"✅ SPACE_HOST found: {space_host_startup}")
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
else:
print("ℹ️ SPACE_HOST environment variable not found (running locally?).")
if space_id_startup: # Print repo URLs if SPACE_ID is found
print(f"✅ SPACE_ID found: {space_id_startup}")
print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}")
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
else:
print("ℹ️ SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")
print("-"*(60 + len(" App Starting ")) + "\n")
print("Launching Gradio Interface for Enhanced Agent with Google Search & Gemini...")
demo.launch(debug=True, share=False)