aiecs 1.3.3__py3-none-any.whl → 1.3.5__py3-none-any.whl

aiecs/__init__.py

@@ -5,7 +5,7 @@ A powerful Python middleware framework for building AI-powered applications
 with tool orchestration, task execution, and multi-provider LLM support.
 """
 
-__version__ = "1.3.3"
+__version__ = "1.3.5"
 __author__ = "AIECS Team"
 __email__ = "iretbl@gmail.com"
 

aiecs/llm/clients/vertex_client.py

@@ -23,6 +23,12 @@ class VertexAIClient(BaseLLMClient):
         super().__init__("Vertex")
         self.settings = get_settings()
         self._initialized = False
+        # Track part count statistics for monitoring
+        self._part_count_stats = {
+            "total_responses": 0,
+            "part_counts": {},  # {part_count: frequency}
+            "last_part_count": None
+        }
 
     def _init_vertex_ai(self):
         """Lazy initialization of Vertex AI with proper authentication"""
@@ -141,19 +147,44 @@ class VertexAIClient(BaseLLMClient):
                                     text_parts.append(part.text)
                             
                             if text_parts:
+                                # Log part count for monitoring
+                                part_count = len(text_parts)
+                                self.logger.info(f"📊 Vertex AI response: {part_count} parts detected")
+                                
+                                # Update statistics
+                                self._part_count_stats["total_responses"] += 1
+                                self._part_count_stats["part_counts"][part_count] = self._part_count_stats["part_counts"].get(part_count, 0) + 1
+                                self._part_count_stats["last_part_count"] = part_count
+                                
+                                # Log statistics if significant variation detected
+                                if part_count != self._part_count_stats.get("last_part_count", part_count):
+                                    self.logger.warning(f"⚠️ Part count variation detected: {part_count} parts (previous: {self._part_count_stats.get('last_part_count', 'unknown')})")
+                                
                                 # Handle multi-part response format
                                 if len(text_parts) > 1:
-                                    # Multi-part response (typical for Gemini 2.5 with tool calling)
-                                    # Part 1: Thinking/reasoning content
-                                    # Part 2+: Actual output (JSON or other format)
+                                    # Multi-part response
+                                    # Minimal fix: only fix incomplete <thinking> tags, preserve original order
+                                    # Do NOT reorganize content - let downstream code handle semantics
                                     
-                                    # Wrap first part (thinking) in <thinking> tags
-                                    thinking_part = text_parts[0]
-                                    actual_output_parts = text_parts[1:]
+                                    processed_parts = []
+                                    fixed_count = 0
                                     
-                                    # Format: <thinking>Part 1</thinking>\nPart 2\nPart 3...
-                                    content = f"<thinking>\n{thinking_part}\n</thinking>\n" + "\n".join(actual_output_parts)
-                                    self.logger.info(f"✅ Successfully wrapped multi-part response: {len(text_parts)} parts (thinking + output)")
+                                    for i, part in enumerate(text_parts):
+                                        if '<thinking>' in part and '</thinking>' not in part:
+                                            # Incomplete thinking tag: add closing tag
+                                            part = part + '\n</thinking>'
+                                            fixed_count += 1
+                                            self.logger.debug(f"  Part {i+1}: Incomplete <thinking> tag fixed")
+                                        
+                                        processed_parts.append(part)
+                                    
+                                    # Merge in original order
+                                    content = "\n".join(processed_parts)
+                                    
+                                    if fixed_count > 0:
+                                        self.logger.info(f"✅ Multi-part response merged: {len(text_parts)} parts, {fixed_count} incomplete tags fixed, order preserved")
+                                    else:
+                                        self.logger.info(f"✅ Multi-part response merged: {len(text_parts)} parts, order preserved")
                                 else:
                                     # Single part response - use as is
                                     content = text_parts[0]
@@ -238,7 +269,64 @@ class VertexAIClient(BaseLLMClient):
             yield word + " "
             await asyncio.sleep(0.05)  # Small delay to simulate streaming
 
+    def get_part_count_stats(self) -> Dict[str, Any]:
+        """
+        Get statistics about part count variations in Vertex AI responses.
+        
+        Returns:
+            Dictionary containing part count statistics and analysis
+        """
+        stats = self._part_count_stats.copy()
+        
+        if stats["total_responses"] > 0:
+            # Calculate variation metrics
+            part_counts = list(stats["part_counts"].keys())
+            stats["variation_analysis"] = {
+                "unique_part_counts": len(part_counts),
+                "most_common_count": max(stats["part_counts"].items(), key=lambda x: x[1])[0] if stats["part_counts"] else None,
+                "part_count_range": f"{min(part_counts)}-{max(part_counts)}" if part_counts else "N/A",
+                "stability_score": 1.0 - (len(part_counts) - 1) / max(stats["total_responses"], 1)  # 0-1, higher is more stable
+            }
+            
+            # Generate recommendations
+            if stats["variation_analysis"]["stability_score"] < 0.7:
+                stats["recommendations"] = [
+                    "High part count variation detected",
+                    "Consider optimizing prompt structure",
+                    "Monitor input complexity patterns",
+                    "Review tool calling configuration"
+                ]
+            else:
+                stats["recommendations"] = [
+                    "Part count variation is within acceptable range",
+                    "Continue monitoring for patterns"
+                ]
+        
+        return stats
+    
+    def log_part_count_summary(self):
+        """Log a summary of part count statistics"""
+        stats = self.get_part_count_stats()
+        
+        if stats["total_responses"] > 0:
+            self.logger.info("📈 Vertex AI Part Count Summary:")
+            self.logger.info(f"  Total responses: {stats['total_responses']}")
+            self.logger.info(f"  Part count distribution: {stats['part_counts']}")
+            
+            if "variation_analysis" in stats:
+                analysis = stats["variation_analysis"]
+                self.logger.info(f"  Stability score: {analysis['stability_score']:.2f}")
+                self.logger.info(f"  Most common count: {analysis['most_common_count']}")
+                self.logger.info(f"  Count range: {analysis['part_count_range']}")
+                
+                if "recommendations" in stats:
+                    self.logger.info("  Recommendations:")
+                    for rec in stats["recommendations"]:
+                        self.logger.info(f"    • {rec}")
+
     async def close(self):
         """Clean up resources"""
+        # Log final statistics before cleanup
+        self.log_part_count_summary()
         # Vertex AI doesn't require explicit cleanup
         self._initialized = False

aiecs/main.py

@@ -142,7 +142,7 @@ async def lifespan(app: FastAPI):
 app = FastAPI(
     title="AIECS - AI Execute Services",
     description="Middleware service for AI-powered task execution and tool orchestration",
-    version="1.3.3",    
+    version="1.3.5",    
     lifespan=lifespan
 )
 
@@ -167,7 +167,7 @@ async def health_check():
     return {
         "status": "healthy",
         "service": "aiecs",
-        "version": "1.3.3"
+        "version": "1.3.5"
     }
 
 

{aiecs-1.3.3.dist-info → aiecs-1.3.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aiecs
-Version: 1.3.3
+Version: 1.3.5
 Summary: AI Execute Services - A middleware framework for AI-powered task execution and tool orchestration
 Author-email: AIECS Team <iretbl@gmail.com>
 License-Expression: MIT

{aiecs-1.3.3.dist-info → aiecs-1.3.5.dist-info}/RECORD

@@ -1,7 +1,7 @@
-aiecs/__init__.py,sha256=_1XRwx-1uwE4PsagfdYcA2kuDIrAOPxmK6OBVtKf_qA,1859
+aiecs/__init__.py,sha256=0C3__5CeIMR8OU-fqcmKw2ETo4SIjYuFNk_MZNI_wfI,1859
 aiecs/__main__.py,sha256=AfQpzy3SgwWuP4DuymYcm4MISMuzqwhxxGSYo53PBvY,1035
 aiecs/aiecs_client.py,sha256=gIqecRBBH_bYIWhqiHCemdVgmGb9Jqdxf1b6RoqXWlQ,17276
-aiecs/main.py,sha256=QF_ln8faRFzoT-kTAC9ZbNVskodN9nkr22ZR_wQZfq8,10837
+aiecs/main.py,sha256=jYY2nqk00XGoEtyQdLNkmpt-4RF52QeQyZUNU0305Ig,10837
 aiecs/application/__init__.py,sha256=NkmrUH1DqxJ3vaVC8QwscNdlWqHfC7ZagL4k3nZ_qz4,192
 aiecs/application/executors/__init__.py,sha256=WIl7L9HBsEhNfbNtJdvBvFUJXzESvNZVaiAA6tdtJcs,191
 aiecs/application/executors/operation_executor.py,sha256=-7mFo1hUnWdehVPg0fnSiRhW3LACpIiyLSH-iu7bX4U,13818
@@ -58,7 +58,7 @@ aiecs/llm/clients/__init__.py,sha256=uQM004TQappwJMqTxVZNscpVPQtirkvYUPea3QYB7d0
 aiecs/llm/clients/base_client.py,sha256=j4NY-oEdG5ALBCSddblPpjttISn5teqLVVUuZyYn7g4,5880
 aiecs/llm/clients/googleai_client.py,sha256=sTgdw4eicxWruNGOMSsuEHbfF9RuDQo8SClqEtu1JOQ,6591
 aiecs/llm/clients/openai_client.py,sha256=x7Y_yTVu0kp-gu5Z-M0Bx-O20D0YDgZoJQxzkjNpr6c,4202
-aiecs/llm/clients/vertex_client.py,sha256=jyyNMeRTJX-QtM95qjor-qA2iEOBO4s7940uv910_J4,12386
+aiecs/llm/clients/vertex_client.py,sha256=Vd5IJIvPVQuv0l6bMQmkY2aemrkfLoHY2RM5Z5SUhVg,17139
 aiecs/llm/clients/xai_client.py,sha256=XEELb9_qFeeQaansDWvAJRJVpt8CaBRLcYskuv9uDq0,6386
 aiecs/llm/config/__init__.py,sha256=KZbwHoBlbcN7HgNueA5p-0GpyVMJRNG1V5T-tkri8G4,1115
 aiecs/llm/config/config_loader.py,sha256=PTMsZax3CoTrMo6BZlUoI7takII3_DHm3w5xTKgBJpA,8921
@@ -168,9 +168,9 @@ aiecs/utils/prompt_loader.py,sha256=cBS2bZXpYQOWSiOGkhwIzyy3_bETqwIblRi_9qQT9iQ,
 aiecs/utils/token_usage_repository.py,sha256=1xjenLYwC0YT6lKZFEGO4scRCXLuWdec2MWjzih5SZY,10210
 aiecs/ws/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 aiecs/ws/socket_server.py,sha256=j_9idVY_rWlTsF51FgmuhWCWFVt7_gAHL8vNg3IxV5g,1476
-aiecs-1.3.3.dist-info/licenses/LICENSE,sha256=_1YRaIS0eZu1pv6xfz245UkU0i1Va2B841hv3OWRwqg,12494
-aiecs-1.3.3.dist-info/METADATA,sha256=5AuGrBEyu3n9K-4iQQUQTfdKsuuBucisZ8ip4JOWYQ4,16635
-aiecs-1.3.3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-aiecs-1.3.3.dist-info/entry_points.txt,sha256=TfLBuwLOfgQqKvnoF1sgTS19-Hgl0aWvCZjIdblIiig,667
-aiecs-1.3.3.dist-info/top_level.txt,sha256=22IlUlOqh9Ni3jXlQNMNUqzbW8dcxXPeR_EQ-BJVcV8,6
-aiecs-1.3.3.dist-info/RECORD,,
+aiecs-1.3.5.dist-info/licenses/LICENSE,sha256=_1YRaIS0eZu1pv6xfz245UkU0i1Va2B841hv3OWRwqg,12494
+aiecs-1.3.5.dist-info/METADATA,sha256=34_y02-Q5OdrWlquol8HuRW0QOgP6WRrSHM9uCaYuQM,16635
+aiecs-1.3.5.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+aiecs-1.3.5.dist-info/entry_points.txt,sha256=TfLBuwLOfgQqKvnoF1sgTS19-Hgl0aWvCZjIdblIiig,667
+aiecs-1.3.5.dist-info/top_level.txt,sha256=22IlUlOqh9Ni3jXlQNMNUqzbW8dcxXPeR_EQ-BJVcV8,6
+aiecs-1.3.5.dist-info/RECORD,,