modifies orchestrator,add code_exec tool, openai_chat (later we add chat)

orchestrator_agent.py

@@ -1,5 +1,4 @@
 
-from datetime import datetime
 import os
 from typing import Dict, List, Any
 from pydantic_ai import Agent
@@ -8,12 +7,10 @@ from pydantic_ai.providers.google_gla import GoogleGLAProvider
 from pydantic_ai import RunContext
 from pydantic import BaseModel
 from google.api_core.exceptions import ResourceExhausted  # Import the exception for quota exhaustion
-from code_exec_service import run_analysis
 from csv_service import get_csv_basic_info
 from orchestrator_functions import csv_chart, csv_chat
 from dotenv import load_dotenv
 
-
 load_dotenv()
 
 
@@ -128,14 +125,14 @@ def create_agent(csv_url: str, api_key: str, conversation_history: List) -> Agen
    - Highlight limitations/caveats
 
 5. TOOL USAGE:
-   - Python Code Executor Tool (To execute Python code, To get date-time, For lightweight data analysis etc.)
-   - Data Analysis Tool
-   - Chart Generation Tool
+   - Can process statistical operations
+   - Supports visualization libraries
 
 ## Current Context:
 - Working with CSV_URL: {csv_url}
 - Dataset overview: {csv_metadata}
 - Your conversation history: {conversation_history}
+- Output format: Markdown compatible
 
 ## Response Template:
 1. Confirm understanding of request
@@ -144,44 +141,58 @@ def create_agent(csv_url: str, api_key: str, conversation_history: List) -> Agen
 4. Provide interpretation
 5. Offer next-step suggestions
 """
-    gemini_csv_orchestrator_agent = Agent(
+    
+    # system_prompt = (
+    # "You are a data analyst. "
+    # "You have all the tools you need to answer any question. "
+    # "If the user asks for multiple answers or charts, break the question into several well-defined questions. "
+    # "Pass the CSV URL or file path along with the questions to the tools to generate the answer. "
+    # "The tools are actually LLMs with Python code execution capabilities. "
+    # "Modify the query if needed to make it simpler for the LLM to understand. "
+    # "Answer in a friendly and helpful manner. "
+    # "**Format images** in Markdown: `![alt_text](direct-image-url)`. "
+    # f"Your CSV URL is {csv_url}. "
+    # f"Your CSV metadata is {csv_metadata}."
+    # )
+
+    
+    # system_prompt = (
+    #     "You are a data analyst assistant with limited tool capabilities. "
+    #     "Available tools can only handle simple data queries: "
+    #     "- Count rows/columns\n- Calculate basic stats (avg, sum, min/max)\n"
+    #     "- Create simple visualizations (pie charts, bar graphs)\n"
+    #     "- Show column names/types\n\n"
+        
+    #     "Query Handling Rules:\n"
+    #     "1. If query is complex, ambiguous, or exceeds tool capabilities:\n"
+    #     "   - Break into simpler sub-questions\n"
+    #     "   - Ask for clarification\n"
+    #     "   - Rephrase to nearest simple query\n"
+    #     "2. For 'full report' requests:\n"
+    #     "   - Outline possible analysis steps\n"
+    #     "   - Ask user to select one component at a time\n\n"
+        
+    #     "Examples:\n"
+    #     "- Bad query: 'Show me everything'\n"
+    #     "  Response: 'I can show row count (10), columns (5: Name, Age...), "
+    #     "or a pie chart of categories. Which would you like?'\n"
+    #     "- Bad query: 'Analyze trends'\n"
+    #     "  Response: 'For trend analysis, I can show monthly averages or "
+    #     "year-over-year comparisons. Please specify time period and metric.'\n\n"
+        
+    #     "Current CSV Context:\n"
+    #     f"- URL: {csv_url}\n"
+    #     f"- Metadata: {csv_metadata}\n\n"
+        
+    #     "Always format images as: ![Chart Description](direct-image-url)"
+    # )
+    
+    return Agent(
         model=initialize_model(api_key),
         deps_type=str,
         tools=[generate_csv_answer, generate_chart],
         system_prompt=system_prompt
     )
-    
-    @gemini_csv_orchestrator_agent.tool_plain
-    def python_code_executor(analysis_code: str) -> dict:
-                """_summary_
-
-                Args:
-                    analysis_code (str): _description_
-                    Ex: 
-                       df = pd.read_csv({csv_url})
-                       len(df)
-
-                Returns:
-                   dict: _description_
-                """
-                
-                print(f'LLM Passed a code: {analysis_code}')
-                result = run_analysis(analysis_code)
-                if result['success']:
-                    print("Execution successful")
-                    print("Execution time:", result['execution_time'], "seconds")
-                    print("Output:", result['output'].strip())
-                    print("Result:", result['result'])
-                    print("Variables:", list(result['variables'].keys()))
-                    # convert the result to a string
-                    result_str = str(result['output'])
-                    return result_str
-                else:
-                    print("Execution failed")
-                    print("Error:", result['error'])
-                    error_str = str(result['error'])
-                    return error_str
-    return gemini_csv_orchestrator_agent
 
 def csv_orchestrator_chat(csv_url: str, user_question: str, conversation_history: List) -> str:
     print("CSV URL:", csv_url)
@@ -205,4 +216,3 @@ def csv_orchestrator_chat(csv_url: str, user_question: str, conversation_history
     # If all keys are exhausted or fail
     print("All API keys have been exhausted or failed.")
     return None
-