Make sure disabling agentic thought only pass MedSwin

pipeline.py

@@ -428,7 +428,10 @@ def stream_chat(
     
     clinical_intake_context_block = ""
     
-    if not enable_clinical_intake:
+    # Clinical intake currently uses Gemini-based supervisors.
+    # When agentic reasoning is disabled, we also skip all Gemini-driven
+    # intake planning and summarization so the flow is purely MedSwin.
+    if disable_agentic_reasoning or not enable_clinical_intake:
         _clear_clinical_intake_state(user_id)
     else:
         intake_state = _rehydrate_intake_state(user_id, history)
@@ -711,67 +714,77 @@ def stream_chat(
             continue
     record_stage("medswin_tasks", medswin_stage_start)
     
-    logger.info("[GEMINI SUPERVISOR] Synthesizing final answer from all MedSwin responses...")
-    raw_medswin_answers = [ans.split('\n\n', 1)[1] if '\n\n' in ans else ans for ans in medswin_answers]
-    synthesis_stage_start = time.time()
-    # Run in thread pool to avoid blocking GPU task
-    final_answer = run_gemini_in_thread(
-        gemini_supervisor_synthesize, message, raw_medswin_answers, rag_contexts, search_contexts, breakdown
-    )
-    record_stage("synthesis", synthesis_stage_start)
-    
-    if not final_answer or len(final_answer.strip()) < 50:
-        logger.warning("[GEMINI SUPERVISOR] Synthesis failed or too short, using concatenation")
-        final_answer = "\n\n".join(medswin_answers) if medswin_answers else "I apologize, but I was unable to generate a response."
-    
-    if "|" in final_answer and "---" in final_answer:
-        logger.warning("[MEDSWIN] Final answer contains tables, converting to bullets")
-        lines = final_answer.split('\n')
-        cleaned_lines = []
-        for line in lines:
-            if '|' in line and '---' not in line:
-                cells = [cell.strip() for cell in line.split('|') if cell.strip()]
-                if cells:
-                    cleaned_lines.append(f"- {' / '.join(cells)}")
-            elif '---' not in line:
-                cleaned_lines.append(line)
-        final_answer = '\n'.join(cleaned_lines)
-    
-    max_challenge_iterations = 2
-    challenge_iteration = 0
-    challenge_stage_start = time.time()
-    
-    while challenge_iteration < max_challenge_iterations and elapsed() < soft_timeout - 15:
-        challenge_iteration += 1
-        logger.info(f"[GEMINI SUPERVISOR] Challenge iteration {challenge_iteration}/{max_challenge_iterations}...")
-        
+    # If agentic reasoning is disabled, we skip all Gemini-based synthesis,
+    # challenge, and enhancement loops. The final answer is just the
+    # concatenation of MedSwin task outputs.
+    if disable_agentic_reasoning:
+        logger.info("[MAC] Agentic reasoning disabled - skipping Gemini synthesis and challenge")
+        if medswin_answers:
+            final_answer = "\n\n".join(medswin_answers)
+        else:
+            final_answer = "I apologize, but I was unable to generate a response."
+    else:
+        logger.info("[GEMINI SUPERVISOR] Synthesizing final answer from all MedSwin responses...")
+        raw_medswin_answers = [ans.split('\n\n', 1)[1] if '\n\n' in ans else ans for ans in medswin_answers]
+        synthesis_stage_start = time.time()
         # Run in thread pool to avoid blocking GPU task
-        evaluation = run_gemini_in_thread(
-            gemini_supervisor_challenge, message, final_answer, raw_medswin_answers, rag_contexts, search_contexts
+        final_answer = run_gemini_in_thread(
+            gemini_supervisor_synthesize, message, raw_medswin_answers, rag_contexts, search_contexts, breakdown
         )
+        record_stage("synthesis", synthesis_stage_start)
         
-        if evaluation.get("is_optimal", False):
-            logger.info(f"[GEMINI SUPERVISOR] Answer confirmed optimal after {challenge_iteration} iteration(s)")
-            break
+        if not final_answer or len(final_answer.strip()) < 50:
+            logger.warning("[GEMINI SUPERVISOR] Synthesis failed or too short, using concatenation")
+            final_answer = "\n\n".join(medswin_answers) if medswin_answers else "I apologize, but I was unable to generate a response."
         
-        enhancement_instructions = evaluation.get("enhancement_instructions", "")
-        if not enhancement_instructions:
-            logger.info("[GEMINI SUPERVISOR] No enhancement instructions, considering answer optimal")
-            break
+        if "|" in final_answer and "---" in final_answer:
+            logger.warning("[MEDSWIN] Final answer contains tables, converting to bullets")
+            lines = final_answer.split('\n')
+            cleaned_lines = []
+            for line in lines:
+                if '|' in line and '---' not in line:
+                    cells = [cell.strip() for cell in line.split('|') if cell.strip()]
+                    if cells:
+                        cleaned_lines.append(f"- {' / '.join(cells)}")
+                elif '---' not in line:
+                    cleaned_lines.append(line)
+            final_answer = '\n'.join(cleaned_lines)
         
-        logger.info(f"[GEMINI SUPERVISOR] Enhancing answer based on feedback...")
-        # Run in thread pool to avoid blocking GPU task
-        enhanced_answer = run_gemini_in_thread(
-            gemini_supervisor_enhance_answer, message, final_answer, enhancement_instructions, raw_medswin_answers, rag_contexts, search_contexts
-        )
+        max_challenge_iterations = 2
+        challenge_iteration = 0
+        challenge_stage_start = time.time()
         
-        if enhanced_answer and len(enhanced_answer.strip()) > len(final_answer.strip()) * 0.8:
-            final_answer = enhanced_answer
-            logger.info(f"[GEMINI SUPERVISOR] Answer enhanced (new length: {len(final_answer)} chars)")
-        else:
-            logger.info("[GEMINI SUPERVISOR] Enhancement did not improve answer significantly, stopping")
-            break
-    record_stage("challenge_loop", challenge_stage_start)
+        while challenge_iteration < max_challenge_iterations and elapsed() < soft_timeout - 15:
+            challenge_iteration += 1
+            logger.info(f"[GEMINI SUPERVISOR] Challenge iteration {challenge_iteration}/{max_challenge_iterations}...")
+            
+            # Run in thread pool to avoid blocking GPU task
+            evaluation = run_gemini_in_thread(
+                gemini_supervisor_challenge, message, final_answer, raw_medswin_answers, rag_contexts, search_contexts
+            )
+            
+            if evaluation.get("is_optimal", False):
+                logger.info(f"[GEMINI SUPERVISOR] Answer confirmed optimal after {challenge_iteration} iteration(s)")
+                break
+            
+            enhancement_instructions = evaluation.get("enhancement_instructions", "")
+            if not enhancement_instructions:
+                logger.info("[GEMINI SUPERVISOR] No enhancement instructions, considering answer optimal")
+                break
+            
+            logger.info(f"[GEMINI SUPERVISOR] Enhancing answer based on feedback...")
+            # Run in thread pool to avoid blocking GPU task
+            enhanced_answer = run_gemini_in_thread(
+                gemini_supervisor_enhance_answer, message, final_answer, enhancement_instructions, raw_medswin_answers, rag_contexts, search_contexts
+            )
+            
+            if enhanced_answer and len(enhanced_answer.strip()) > len(final_answer.strip()) * 0.8:
+                final_answer = enhanced_answer
+                logger.info(f"[GEMINI SUPERVISOR] Answer enhanced (new length: {len(final_answer)} chars)")
+            else:
+                logger.info("[GEMINI SUPERVISOR] Enhancement did not improve answer significantly, stopping")
+                break
+        record_stage("challenge_loop", challenge_stage_start)
     
     if final_use_web_search and elapsed() < soft_timeout - 10:
         logger.info("[GEMINI SUPERVISOR] Checking if additional search is needed...")