import os import io import sys import re import traceback import subprocess import gradio as gr import pandas as pd from dotenv import load_dotenv from crewai import Crew, Agent, Task, Process, LLM from crewai_tools import FileReadTool from pydantic import BaseModel, Field # Load environment variables load_dotenv() # Get API key from environment variables OPENAI_API_KEY = os.getenv('OPENAI_API_KEY') if not OPENAI_API_KEY: raise ValueError("OPENAI_API_KEY environment variable not set") llm = LLM( model="openai/gpt-4o", api_key=OPENAI_API_KEY, temperature=0.7 ) # 1) Query parser agent query_parser_agent = Agent( role="Stock Data Analyst", goal="Extract stock details and fetch required data from this user query: {query}.", backstory="You are a financial analyst specializing in stock market data retrieval.", llm=llm, verbose=True, memory=True, ) # Need to define QueryAnalysisOutput class here as it's used by the task class QueryAnalysisOutput(BaseModel): """Structured output for the query analysis task.""" symbols: list[str] = Field( ..., json_schema_extra={"description": "List of stock ticker symbols (e.g., ['TSLA', 'AAPL'])."} ) timeframe: str = Field( ..., json_schema_extra={"description": "Time period (e.g., '1d', '1mo', '1y')."} ) action: str = Field( ..., json_schema_extra={"description": "Action to be performed (e.g., 'fetch', 'plot')."} ) query_parsing_task = Task( description="Analyze the user query and extract stock details.", expected_output="A dictionary with keys: 'symbol', 'timeframe', 'action'.", output_pydantic=QueryAnalysisOutput, agent=query_parser_agent, ) # 2) Code writer agent code_writer_agent = Agent( role="Senior Python Developer", goal="Write Python code to visualize stock data.", backstory="""You are a Senior Python developer specializing in stock market data visualization. You are also a Pandas, Matplotlib and yfinance library expert. You are skilled at writing production-ready Python code. Ensure the code handles potential variations in the DataFrame structure returned by yfinance, especially for different timeframes or delisted stocks. Crucially, ensure the generated script saves any generated plot as 'plot.png' using `plt.savefig('plot.png')` before the script ends.""", llm=llm, verbose=True, ) code_writer_task = Task( description="""Write Python code to visualize stock data based on the inputs from the stock analyst where you would find stock symbol, timeframe and action.""", expected_output="A clean and executable Python script file (.py) for stock visualization.", agent=code_writer_agent, ) # 3) Code output agent (instead of execution agent) code_output_agent = Agent( role="Python Code Presenter", goal="Present the generated Python code for stock visualization.", backstory="You are an expert in presenting Python code in a clear and readable format.", allow_delegation=False, # This agent just presents the code llm=llm, verbose=True, ) code_output_task = Task( description="""Receive the Python code for stock visualization from the code writer agent and present it.""", expected_output="The complete Python script for stock visualization.", agent=code_output_agent, ) crew = Crew( agents=[query_parser_agent, code_writer_agent, code_output_agent], # Use code_output_agent tasks=[query_parsing_task, code_writer_task, code_output_task], # Use code_output_task process=Process.sequential ) def run_crewai_process(user_query, model, temperature): """ Runs the CrewAI process, captures agent thoughts, gets generated code, executes the code, and returns results, including plot. Args: user_query (str): The user's query for the CrewAI process. model (str): The model to use for the LLM. temperature (float): The temperature to use for the LLM. Yields: tuple: A tuple containing the agent thoughts (str), the final answer (list of dicts), the generated code (str), the execution output (str), and plot file path (str or None). """ # Create a string buffer to capture stdout output_buffer = io.StringIO() original_stdout = sys.stdout sys.stdout = output_buffer agent_thoughts = "" generated_code = "" execution_output = "" generated_plot_path = None final_answer_chat = [{"role": "user", "content": user_query}] try: # Kick off the crew process # CrewAI's kickoff doesn't directly support streaming to a buffer # We'll run it and then capture the full output at the end of the CrewAI process # However, for demonstration, we can yield intermediate status updates. yield "Starting CrewAI process...", final_answer_chat, agent_thoughts, generated_code, execution_output, generated_plot_path final_result = crew.kickoff(inputs={"query": user_query}) # Get the captured CrewAI output (agent thoughts) agent_thoughts = output_buffer.getvalue() yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path # The final result is the generated code from the code_output_agent generated_code_raw = str(final_result).strip() # Use regex to extract the code block code_match = re.search(r"```python\n(.*?)\n```", generated_code_raw, re.DOTALL) if code_match: generated_code = code_match.group(1).strip() else: # If no code block is found, assume the entire output is code (or handle as error) generated_code = generated_code_raw if not generated_code.strip(): # Handle cases where output is empty or just whitespace execution_output = "CrewAI process completed, but no code was generated." final_answer_chat.append({"role": "assistant", "content": execution_output}) yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path return # Exit the generator # Format for Gradio Chatbot (list of dictionaries with 'role' and 'content' keys only) final_answer_chat = [ {"role": "user", "content": str(user_query)}, {"role": "assistant", "content": "Code generation complete. See the 'Generated Code' box. Attempting to execute code..."} ] yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path # --- Execute the generated code --- plot_file_path = 'plot.png' # Expected plot file name if generated_code: try: # Write the generated code to a temporary file temp_script_path = "generated_script.py" with open(temp_script_path, "w") as f: f.write(generated_code) # Execute the temporary script using subprocess # Use python3 to ensure correct interpreter in Colab process = subprocess.run( ["python3", temp_script_path], capture_output=True, text=True, # Capture stdout and stderr as text check=False # Don't raise exception for non-zero exit codes ) execution_output = process.stdout + process.stderr # Check for specific errors in execution output if "KeyError" in execution_output: execution_output += "\n\nPotential Issue: The generated script encountered a KeyError. This might mean the script tried to access a column or data point that wasn't available for the specified stock or timeframe. Please try a different query or timeframe." elif "FileNotFoundError: [Errno 2] No such file or directory: 'plot.png'" in execution_output and "Figure(" in execution_output: execution_output += "\n\nPlot Generation Issue: The script seems to have created a plot but did not save it to 'plot.png'. Please ensure the generated code includes `plt.savefig('plot.png')`." elif "FileNotFoundError: [Errno 2] No such file or directory: 'plot.png'" in execution_output: execution_output += "\n\nPlot Generation Issue: The script ran, but the plot file was not created. Ensure the generated code includes commands to save the plot to 'plot.png'." # Check for the generated plot file if os.path.exists(plot_file_path): print(f"Plot file found at: {os.path.abspath(plot_file_path)}") # Log file path generated_plot_path = plot_file_path # Set the path to be returned else: print(f"Plot file not found at expected path: {os.path.abspath(plot_file_path)}") # Log missing file path execution_output += f"\nPlot file '{plot_file_path}' not found after execution." except Exception as e: traceback_str = traceback.format_exc() execution_output = f"An error occurred during code execution: {e}\n{traceback_str}" finally: # Clean up the temporary script file if os.path.exists(temp_script_path): os.remove(temp_script_path) else: execution_output = "No code was generated to execute." # Update final answer chat to reflect execution attempt final_answer_chat = [ {"role": "user", "content": str(user_query)}, {"role": "assistant", "content": "Code execution finished. See 'Execution Output'."} ] if generated_plot_path: final_answer_chat.append({"role": "assistant", "content": "Plot generated successfully. See 'Generated Plot'."}) else: final_answer_chat.append({"role": "assistant", "content": "No plot was generated. Check the execution output for details."}) yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path except Exception as e: # If an error occurs during CrewAI process, return the error message traceback_str = traceback.format_exc() agent_thoughts += f"\nAn error occurred during CrewAI process: {e}\n{traceback_str}" error_message = f"An error occurred during CrewAI process: {e}" final_answer_chat = [ {"role": "user", "content": str(user_query)}, {"role": "assistant", "content": error_message} ] yield agent_thoughts, final_answer_chat, generated_code, execution_output, generated_plot_path finally: # Restore original stdout sys.stdout = original_stdout def create_interface(): """Create and return the Gradio interface.""" with gr.Blocks(title="Financial Analytics Agent", theme=gr.themes.Soft()) as interface: gr.Markdown("# 📊 Financial Analytics Agent") gr.Markdown("Enter your financial query to analyze stock data and generate visualizations.") with gr.Row(): with gr.Column(scale=2): user_query_input = gr.Textbox( label="Enter your financial query", placeholder="e.g., Show me the stock performance of AAPL and MSFT for the last year", lines=3 ) submit_btn = gr.Button("Analyze", variant="primary") with gr.Accordion("Advanced Options", open=False): gr.Markdown("### Model Settings") model_dropdown = gr.Dropdown( ["gpt-4o", "gpt-4-turbo", "gpt-3.5-turbo"], value="gpt-4o", label="Model" ) temperature = gr.Slider( minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Creativity (Temperature)" ) with gr.Column(scale=3): with gr.Tabs(): with gr.TabItem("Analysis"): final_answer_chat = gr.Chatbot( label="Analysis Results", height=300, show_copy_button=True, type="messages" # Explicitly set to use OpenAI-style message format ) with gr.TabItem("Agent Thoughts"): agent_thoughts = gr.Textbox( label="Agent Thinking Process", interactive=False, lines=15, max_lines=30, show_copy_button=True ) with gr.TabItem("Generated Code"): generated_code = gr.Code( label="Generated Python Code", language="python", interactive=False, lines=15 ) with gr.TabItem("Execution Output"): execution_output = gr.Textbox( label="Code Execution Output", interactive=False, lines=10, show_copy_button=True ) with gr.Row(): with gr.Column(): plot_output = gr.Plot( label="Generated Visualization", visible=False ) image_output = gr.Image( label="Generated Plot", type="filepath", visible=False ) # Handle form submission inputs = [user_query_input, model_dropdown, temperature] outputs = [ final_answer_chat, agent_thoughts, generated_code, execution_output, plot_output, image_output ] submit_btn.click( fn=run_crewai_process, inputs=inputs, outputs=outputs, api_name="analyze" ) return interface def main(): """Run the Gradio interface.""" interface = create_interface() interface.launch(share=False, server_name="0.0.0.0", server_port=7860) if __name__ == "__main__": main()