Update app.py

app.py

@@ -151,7 +151,7 @@ Recent History:
 Available Actions and their descriptions:
 {tool_descriptions_for_prompt}
 
-Based on the query and the action descriptions, select the single best action. Output the corresponding JSON.
+Based on the query and the action descriptions, select the single best action to take. Output the corresponding JSON.
 Example for web search: {{"action": "search_duckduckgo_and_report", "action_input": {{"search_engine_query": "latest AI research"}}}}
 Example for memory recall: {{"action": "answer_using_conversation_memory", "action_input": {{}}}}
 """
@@ -190,33 +190,28 @@ Example for memory recall: {{"action": "answer_using_conversation_memory", "acti
         final_user_prompt_content_str = f"History:\n{history_str_for_prompt}\nGuidelines:\n{initial_insights_ctx_str}\nQuery: \"{user_input}\"\nResponse:"
     
     elif action_type == "answer_using_conversation_memory":
-        # --- MODIFIED: Query Transformation Step ---
         yield "status", "<i>[Optimizing query for long-term memory search...]</i>"
         
-        # 1. Ask an LLM to generate a better search query
-        optimized_query = user_input # Fallback
+        optimized_query = user_input
         try:
             query_gen_system_prompt = "You are an expert at reformulating a user's question into a concise, effective search query for a vector database. Given the conversation history and the user's latest query, extract the key topics and entities to create a self-contained query. The query should be a short phrase or question. Output ONLY the query text, nothing else."
             query_gen_user_prompt = f"Conversation History:\n{history_str_for_prompt}\n\nUser's Latest Query: \"{user_input}\"\n\nOptimized Search Query:"
             query_gen_messages = [{"role":"system", "content":query_gen_system_prompt}, {"role":"user", "content":query_gen_user_prompt}]
             
-            # Use the same fast model as the tool decider
             query_gen_chunks = list(call_model_stream(provider=tool_provider, model_display_name=tool_model_display, messages=query_gen_messages, temperature=0.0, max_tokens=50))
             generated_query = "".join(query_gen_chunks).strip()
 
             if generated_query:
-                optimized_query = generated_query.replace('"', '') # Clean up quotes
+                optimized_query = generated_query.replace('"', '')
                 logger.info(f"PUI_GRADIO [{request_id}]: Original query: '{user_input}'. Optimized memory search query: '{optimized_query}'")
             else:
                 logger.warning(f"PUI_GRADIO [{request_id}]: Query generation returned empty. Using original input.")
         except Exception as e:
             logger.error(f"PUI_GRADIO [{request_id}]: Error during query generation: {e}. Using original input.")
 
-        # 2. Use the optimized query to search the unified memory
         yield "status", f"<i>[Searching all memories with query: '{optimized_query[:40]}...']</i>"
         retrieved_mems = retrieve_memories_semantic(optimized_query, k=3)
         
-        # 3. Build the context for the final response
         if retrieved_mems:
             logger.info(f"PUI_GRADIO [{request_id}]: Found {len(retrieved_mems)} relevant memories.")
             memory_context = "Relevant Past Interactions (for your context only, do not repeat verbatim):\n" + "\n".join([f"- User asked: '{m.get('user_input','')}'. You responded: '{m.get('bot_response','')}'. (Key takeaway: {m.get('metrics',{}).get('takeaway','N/A')})" for m in retrieved_mems])
@@ -226,7 +221,6 @@ Example for memory recall: {{"action": "answer_using_conversation_memory", "acti
         
         final_system_prompt_str += " You MUST use the provided 'Memory Context' to inform your answer. Synthesize the information from the memory with the current conversation history to respond to the user's query."
         final_user_prompt_content_str = f"History:\n{history_str_for_prompt}\n\nGuidelines:\n{initial_insights_ctx_str}\n\nMemory Context:\n{memory_context}\n\nUser's Query: \"{user_input}\"\n\nResponse (use the Memory Context to answer the query):"
-        # --- END OF MODIFICATION ---
 
     elif action_type in ["search_duckduckgo_and_report", "scrape_url_and_report"]:
         query_or_url = action_input_dict.get("search_engine_query") if "search" in action_type else action_input_dict.get("url")
@@ -264,7 +258,8 @@ Example for memory recall: {{"action": "answer_using_conversation_memory", "acti
     logger.info(f"PUI_GRADIO [{request_id}]: Finished. Total: {time.time() - process_start_time:.2f}s. Resp len: {len(final_bot_text)}")
     yield "final_response_and_insights", {"response": final_bot_text, "insights_used": parsed_initial_insights_list}
 
-# The rest of the app.py file remains the same...
+# The rest of the app.py file is unchanged and correct.
+# ... (perform_post_interaction_learning, handle_gradio_chat_submit, UI definitions, etc.)
 def perform_post_interaction_learning(user_input: str, bot_response: str, provider: str, model_disp_name: str, insights_reflected: list[dict], api_key_override: str = None):
     task_id = os.urandom(4).hex()
     logger.info(f"POST_INTERACTION_LEARNING [{task_id}]: START User='{user_input[:40]}...', Bot='{bot_response[:40]}...'")