# Importing necessary modules and tools for our application. # These libraries help us work with code, dates, files, images, and the user interface. from smolagents import CodeAgent, DuckDuckGoSearchTool, HfApiModel, load_tool, tool import datetime # Helps with dates and times. import requests # Lets us make web requests. import yaml # Reads YAML files, which are used for configuration. from tools.final_answer import FinalAnswerTool # Custom tool to finalize the answer. from Gradio_UI import GradioUI # Module for creating a simple web interface. # ----------------------------- # Tool: comment_code # ----------------------------- # This function takes a piece of code and adds comments to explain what each part does. @tool def comment_code(code: str) -> str: """ A tool that takes a block of code and returns the same code with added comments. The comments explain each part of the code in simple terms. Args: code: The input code as a string. """ try: # Start with an empty string to collect our commented code. commented_code = "" # Split the code into lines so we can process each line separately. for line in code.split('\n'): stripped = line.strip() # Remove extra spaces from the beginning and end. if stripped: # Check if the line defines a function. if "def " in stripped: commented_code += "# This defines a function\n" # Check if the line imports a module. elif "import " in stripped: commented_code += "# Importing necessary modules\n" # Check if the line is a loop (for or while). elif "for " in stripped or "while " in stripped: commented_code += "# Looping through values\n" # Check if the line contains a conditional statement. elif "if " in stripped: commented_code += "# Checking a condition\n" # Check if the line returns a value. elif "return " in stripped: commented_code += "# Returning a value from the function\n" # Add the original code line after the comment. commented_code += line + "\n" return commented_code except Exception as e: # If something goes wrong, return an error message. return f"Error processing code: {str(e)}" # ----------------------------- # Tool: extract_code_from_image # ----------------------------- # This tool extracts text (code) from an image file, like a screenshot. @tool def extract_code_from_image(image_path: str) -> str: """ Extracts text from an image file (a screenshot) that contains code. Args: image_path: The path to the image file. """ try: from PIL import Image # Module for opening and working with images. import pytesseract # OCR tool to extract text from images. image = Image.open(image_path) # Open the image file. code_text = pytesseract.image_to_string(image) # Extract text from the image. return code_text except Exception as e: # Return an error message if extraction fails. return f"Error extracting code from image: {str(e)}" # ----------------------------- # Tool: extract_code_from_file # ----------------------------- # This tool reads code from a file and returns it as text. @tool def extract_code_from_file(file_path: str) -> str: """ Reads code text from a file. Args: file_path: The path to the file containing code. """ try: with open(file_path, 'r') as f: # Open the file in read mode. code_text = f.read() # Read the content of the file. return code_text except Exception as e: # Return an error message if the file cannot be read. return f"Error reading code from file: {str(e)}" # Create an instance of the FinalAnswerTool. final_answer = FinalAnswerTool() # ----------------------------- # Set up the language model. # ----------------------------- # This model will help the agent understand and process code-related tasks. model = HfApiModel( max_tokens=2096, # Maximum tokens the model can output. temperature=0.5, # Controls randomness in the model's output. model_id='Qwen/Qwen2.5-Coder-32B-Instruct', # Specific model to use. custom_role_conversions=None, # No custom role conversion needed here. ) # ----------------------------- # Optionally, load an image generation tool. # ----------------------------- # This tool is loaded from an external source, though it isn't used directly here. image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True) # ----------------------------- # Load prompt templates from a configuration file. # ----------------------------- # This file likely contains text instructions that help guide the agent. with open("prompts.yaml", 'r') as stream: prompt_templates = yaml.safe_load(stream) # ----------------------------- # Set up the CodeAgent. # ----------------------------- # The agent uses the model and the tools to perform tasks. agent = CodeAgent( model=model, tools=[final_answer, comment_code, extract_code_from_image, extract_code_from_file], max_steps=6, # Limit on how many steps the agent can take. verbosity_level=1, # Level of detail in the logs/output. grammar=None, # No special grammar rules are applied. planning_interval=None, # No specific planning interval set. name="Code Commenter", # Name of the agent. description="An agent that adds explanatory comments to code in simple terms.", prompt_templates=prompt_templates # Prompts to guide the agent's responses. ) # ----------------------------- # Launch the Gradio user interface. # ----------------------------- # Gradio creates a simple web interface for users to interact with the agent. GradioUI(agent).launch()