praisonaiagents 0.0.120__py3-none-any.whl → 0.0.122__py3-none-any.whl

praisonaiagents/agent/agent.py

@@ -1274,27 +1274,40 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                     messages.append({"role": "user", "content": reflection_prompt})
 
                     try:
-                        # Check if OpenAI client is available
-                        if self._openai_client is None:
-                            # For custom LLMs, self-reflection with structured output is not supported
-                            if self.verbose:
-                                display_self_reflection(f"Agent {self.name}: Self-reflection with structured output is not supported for custom LLM providers. Skipping reflection.", console=self.console)
-                            # Return the original response without reflection
-                            self.chat_history.append({"role": "user", "content": prompt})
-                            self.chat_history.append({"role": "assistant", "content": response_text})
-                            # Only display interaction if not using custom LLM (to avoid double output) and verbose is True
-                            if self.verbose and not self._using_custom_llm:
-                                display_interaction(prompt, response_text, markdown=self.markdown, generation_time=time.time() - start_time, console=self.console)
-                            return response_text
-                        
-                        reflection_response = self._openai_client.sync_client.beta.chat.completions.parse(
-                            model=self.reflect_llm if self.reflect_llm else self.llm,
-                            messages=messages,
-                            temperature=temperature,
-                            response_format=ReflectionOutput
-                        )
+                        # Check if we're using a custom LLM (like Gemini)
+                        if self._using_custom_llm or self._openai_client is None:
+                            # For custom LLMs, we need to handle reflection differently
+                            # Use non-streaming to get complete JSON response
+                            reflection_response = self._chat_completion(messages, temperature=temperature, tools=None, stream=False, reasoning_steps=False)
+                            
+                            if not reflection_response or not reflection_response.choices:
+                                raise Exception("No response from reflection request")
+                            
+                            reflection_text = reflection_response.choices[0].message.content.strip()
+                            
+                            # Clean the JSON output
+                            cleaned_json = self.clean_json_output(reflection_text)
+                            
+                            # Parse the JSON manually
+                            reflection_data = json.loads(cleaned_json)
+                            
+                            # Create a reflection output object manually
+                            class CustomReflectionOutput:
+                                def __init__(self, data):
+                                    self.reflection = data.get('reflection', '')
+                                    self.satisfactory = data.get('satisfactory', 'no').lower()
+                            
+                            reflection_output = CustomReflectionOutput(reflection_data)
+                        else:
+                            # Use OpenAI's structured output for OpenAI models
+                            reflection_response = self._openai_client.sync_client.beta.chat.completions.parse(
+                                model=self.reflect_llm if self.reflect_llm else self.llm,
+                                messages=messages,
+                                temperature=temperature,
+                                response_format=ReflectionOutput
+                            )
 
-                        reflection_output = reflection_response.choices[0].message.parsed
+                            reflection_output = reflection_response.choices[0].message.parsed
 
                         if self.verbose:
                             display_self_reflection(f"Agent {self.name} self reflection (using {self.reflect_llm if self.reflect_llm else self.llm}): reflection='{reflection_output.reflection}' satisfactory='{reflection_output.satisfactory}'", console=self.console)
@@ -1337,7 +1350,9 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
 
                         logging.debug(f"{self.name} reflection count {reflection_count + 1}, continuing reflection process")
                         messages.append({"role": "user", "content": "Now regenerate your response using the reflection you made"})
-                        response = self._chat_completion(messages, temperature=temperature, tools=None, stream=self.stream)
+                        # For custom LLMs during reflection, always use non-streaming to ensure complete responses
+                        use_stream = self.stream if not self._using_custom_llm else False
+                        response = self._chat_completion(messages, temperature=temperature, tools=None, stream=use_stream)
                         response_text = response.choices[0].message.content.strip()
                         reflection_count += 1
                         continue  # Continue the loop for more reflections

praisonaiagents/agents/autoagents.py

@@ -13,7 +13,7 @@ import logging
 import os
 from pydantic import BaseModel, ConfigDict
 from ..main import display_instruction, display_tool_call, display_interaction
-from ..llm import get_openai_client, LLM
+from ..llm import get_openai_client, LLM, OpenAIClient
 import json
 
 # Define Pydantic models for structured output
@@ -109,6 +109,8 @@ class AutoAgents(PraisonAIAgents):
         self.max_execution_time = max_execution_time
         self.max_iter = max_iter
         self.reflect_llm = reflect_llm
+        self.base_url = base_url
+        self.api_key = api_key
         
         # Display initial instruction
         if self.verbose:
@@ -212,9 +214,37 @@ Tools: {', '.join(agent_tools)}"""
             
         return assigned_tools
 
+    def _validate_config(self, config: AutoAgentsConfig) -> tuple[bool, str]:
+        """
+        Validate that the configuration has proper TaskConfig objects.
+        
+        Returns:
+            Tuple of (is_valid, error_message)
+        """
+        for agent_idx, agent in enumerate(config.agents):
+            if not hasattr(agent, 'tasks') or not agent.tasks:
+                return False, f"Agent '{agent.name}' has no tasks defined"
+            
+            for task_idx, task in enumerate(agent.tasks):
+                # Check if task is a proper TaskConfig instance
+                if not isinstance(task, TaskConfig):
+                    return False, f"Task at index {task_idx} for agent '{agent.name}' is not a proper TaskConfig object"
+                
+                # Check required fields
+                if not task.name:
+                    return False, f"Task at index {task_idx} for agent '{agent.name}' has no name"
+                if not task.description:
+                    return False, f"Task at index {task_idx} for agent '{agent.name}' has no description"
+                if not task.expected_output:
+                    return False, f"Task at index {task_idx} for agent '{agent.name}' has no expected_output"
+                if task.tools is None:
+                    return False, f"Task at index {task_idx} for agent '{agent.name}' has no tools field"
+        
+        return True, ""
+
     def _generate_config(self) -> AutoAgentsConfig:
-        """Generate the configuration for agents and tasks"""
-        prompt = f"""
+        """Generate the configuration for agents and tasks with retry logic"""
+        base_prompt = f"""
 Generate a configuration for AI agents to accomplish this task: "{self.instructions}"
 
 The configuration should include:
@@ -235,81 +265,164 @@ Requirements:
 4. The process type should match the task requirements
 5. Generate maximum {self.max_agents} agents to handle this task efficiently
 
-Return the configuration in a structured JSON format matching the AutoAgentsConfig schema.
+Return the configuration in a structured JSON format matching this exact schema:
+{{
+  "main_instruction": "Overall goal description",
+  "process_type": "sequential|workflow|hierarchical",
+  "agents": [
+    {{
+      "name": "Agent Name",
+      "role": "Agent Role",
+      "goal": "Agent Goal",
+      "backstory": "Agent Backstory",
+      "tools": ["tool1", "tool2"],
+      "tasks": [
+        {{
+          "name": "Task Name",
+          "description": "Detailed task description",
+          "expected_output": "What the task should produce",
+          "tools": ["tool1", "tool2"]
+        }}
+      ]
+    }}
+  ]
+}}
+
+IMPORTANT: Each task MUST be an object with name, description, expected_output, and tools fields, NOT a simple string.
 """
         
-        try:
-            # Try to use OpenAI's structured output if available
-            use_openai_structured = False
-            client = None
+        max_retries = 3
+        last_response = None
+        last_error = None
+        
+        for attempt in range(max_retries):
+            # Prepare prompt for this attempt
+            if attempt > 0 and last_response and last_error:
+                # On retry, include the previous response and error
+                prompt = f"""{base_prompt}
+
+PREVIOUS ATTEMPT FAILED!
+Your previous response was:
+```json
+{last_response}
+```
+
+Error: {last_error}
+
+REMEMBER: Tasks MUST be objects with the following structure:
+{{
+  "name": "Task Name",
+  "description": "Task Description",
+  "expected_output": "Expected Output",
+  "tools": ["tool1", "tool2"]
+}}
+
+DO NOT use strings for tasks. Each task MUST be a complete object with all four fields."""
+            else:
+                prompt = base_prompt
             
             try:
-                # Check if we have OpenAI API and the model supports structured output
-                if self.llm and (self.llm.startswith('gpt-') or self.llm.startswith('o1-') or self.llm.startswith('o3-')):
-                    client = get_openai_client()
-                    use_openai_structured = True
-            except:
-                # If OpenAI client is not available, we'll use the LLM class
-                pass
-            
-            if use_openai_structured and client:
-                # Use OpenAI's structured output for OpenAI models (backward compatibility)
-                response = client.beta.chat.completions.parse(
-                    model=self.llm,
-                    response_format=AutoAgentsConfig,
-                    messages=[
-                        {"role": "system", "content": "You are a helpful assistant designed to generate AI agent configurations."},
-                        {"role": "user", "content": prompt}
-                    ]
-                )
-                config = response.choices[0].message.parsed
-            else:
-                # Use LLM class for all other providers (Gemini, Anthropic, etc.)
-                llm_instance = LLM(
-                    model=self.llm,
-                    base_url=self.base_url,
-                    api_key=self.api_key
-                )
+                # Try to use OpenAI's structured output if available
+                use_openai_structured = False
+                client = None
                 
-                response_text = llm_instance.response(
-                    prompt=prompt,
-                    system_prompt="You are a helpful assistant designed to generate AI agent configurations.",
-                    output_pydantic=AutoAgentsConfig,
-                    temperature=0.7,
-                    stream=False,
-                    verbose=False
-                )
-                
-                # Parse the JSON response
                 try:
-                    # First try to parse as is
-                    config_dict = json.loads(response_text)
-                    config = AutoAgentsConfig(**config_dict)
-                except json.JSONDecodeError:
-                    # If that fails, try to extract JSON from the response
-                    # Handle cases where the model might wrap JSON in markdown blocks
-                    cleaned_response = response_text.strip()
-                    if cleaned_response.startswith("```json"):
-                        cleaned_response = cleaned_response[7:]
-                    if cleaned_response.startswith("```"):
-                        cleaned_response = cleaned_response[3:]
-                    if cleaned_response.endswith("```"):
-                        cleaned_response = cleaned_response[:-3]
-                    cleaned_response = cleaned_response.strip()
+                    # Check if we have OpenAI API and the model supports structured output
+                    if self.llm and (self.llm.startswith('gpt-') or self.llm.startswith('o1-') or self.llm.startswith('o3-')):
+                        # Create a new client instance if custom parameters are provided
+                        if self.api_key or self.base_url:
+                            client = OpenAIClient(api_key=self.api_key, base_url=self.base_url)
+                        else:
+                            client = get_openai_client()
+                        use_openai_structured = True
+                except:
+                    # If OpenAI client is not available, we'll use the LLM class
+                    pass
+                
+                if use_openai_structured and client:
+                    # Use OpenAI's structured output for OpenAI models (backward compatibility)
+                    config = client.parse_structured_output(
+                        messages=[
+                            {"role": "system", "content": "You are a helpful assistant designed to generate AI agent configurations."},
+                            {"role": "user", "content": prompt}
+                        ],
+                        response_format=AutoAgentsConfig,
+                        model=self.llm
+                    )
+                    # Store the response for potential retry
+                    last_response = json.dumps(config.model_dump(), indent=2)
+                else:
+                    # Use LLM class for all other providers (Gemini, Anthropic, etc.)
+                    llm_instance = LLM(
+                        model=self.llm,
+                        base_url=self.base_url,
+                        api_key=self.api_key
+                    )
                     
-                    config_dict = json.loads(cleaned_response)
-                    config = AutoAgentsConfig(**config_dict)
-            
-            # Ensure we have exactly max_agents number of agents
-            if len(config.agents) > self.max_agents:
-                config.agents = config.agents[:self.max_agents]
-            elif len(config.agents) < self.max_agents:
-                logging.warning(f"Generated {len(config.agents)} agents, expected {self.max_agents}")
-            
-            return config
-        except Exception as e:
-            logging.error(f"Error generating configuration: {e}")
-            raise
+                    response_text = llm_instance.response(
+                        prompt=prompt,
+                        system_prompt="You are a helpful assistant designed to generate AI agent configurations.",
+                        output_pydantic=AutoAgentsConfig,
+                        temperature=0.7,
+                        stream=False,
+                        verbose=False
+                    )
+                    
+                    # Store the raw response for potential retry
+                    last_response = response_text
+                    
+                    # Parse the JSON response
+                    try:
+                        # First try to parse as is
+                        config_dict = json.loads(response_text)
+                        config = AutoAgentsConfig(**config_dict)
+                    except json.JSONDecodeError:
+                        # If that fails, try to extract JSON from the response
+                        # Handle cases where the model might wrap JSON in markdown blocks
+                        cleaned_response = response_text.strip()
+                        if cleaned_response.startswith("```json"):
+                            cleaned_response = cleaned_response[7:]
+                        if cleaned_response.startswith("```"):
+                            cleaned_response = cleaned_response[3:]
+                        if cleaned_response.endswith("```"):
+                            cleaned_response = cleaned_response[:-3]
+                        cleaned_response = cleaned_response.strip()
+                        
+                        config_dict = json.loads(cleaned_response)
+                        config = AutoAgentsConfig(**config_dict)
+                
+                # Validate the configuration
+                is_valid, error_msg = self._validate_config(config)
+                if not is_valid:
+                    last_error = error_msg
+                    if attempt < max_retries - 1:
+                        logging.warning(f"Configuration validation failed (attempt {attempt + 1}/{max_retries}): {error_msg}")
+                        continue
+                    else:
+                        raise ValueError(f"Configuration validation failed after {max_retries} attempts: {error_msg}")
+                
+                # Ensure we have exactly max_agents number of agents
+                if len(config.agents) > self.max_agents:
+                    config.agents = config.agents[:self.max_agents]
+                elif len(config.agents) < self.max_agents:
+                    logging.warning(f"Generated {len(config.agents)} agents, expected {self.max_agents}")
+                
+                return config
+                
+            except ValueError as e:
+                # Re-raise validation errors
+                raise
+            except Exception as e:
+                last_error = str(e)
+                if attempt < max_retries - 1:
+                    logging.warning(f"Error generating configuration (attempt {attempt + 1}/{max_retries}): {e}")
+                    continue
+                else:
+                    logging.error(f"Error generating configuration after {max_retries} attempts: {e}")
+                    raise
+        
+        # This should never be reached due to the raise statements above
+        raise RuntimeError(f"Failed to generate valid configuration after {max_retries} attempts")
 
     def _create_agents_and_tasks(self, config: AutoAgentsConfig) -> Tuple[List[Agent], List[Task]]:
         """Create agents and tasks from configuration"""
@@ -350,7 +463,9 @@ Return the configuration in a structured JSON format matching the AutoAgentsConf
                 max_rpm=self.max_rpm,
                 max_execution_time=self.max_execution_time,
                 max_iter=self.max_iter,
-                reflect_llm=self.reflect_llm
+                reflect_llm=self.reflect_llm,
+                base_url=self.base_url,
+                api_key=self.api_key
             )
             agents.append(agent)
             

praisonaiagents/llm/llm.py

@@ -746,6 +746,7 @@ class LLM:
                                             )
                                             if delta.content:
                                                 live.update(display_generating(response_text, current_time))
+
                             else:
                                 # Non-verbose streaming
                                 for chunk in litellm.completion(
@@ -759,9 +760,12 @@ class LLM:
                                 ):
                                     if chunk and chunk.choices and chunk.choices[0].delta:
                                         delta = chunk.choices[0].delta
-                                        response_text, tool_calls = self._process_stream_delta(
-                                            delta, response_text, tool_calls, formatted_tools
-                                        )
+                                        if delta.content:
+                                            response_text += delta.content
+                                        
+                                        # Capture tool calls from streaming chunks if provider supports it
+                                        if formatted_tools and self._supports_streaming_tools():
+                                            tool_calls = self._process_tool_calls_from_stream(delta, tool_calls)
                             
                             response_text = response_text.strip()
                             
@@ -802,20 +806,7 @@ class LLM:
                     # Handle tool calls - Sequential tool calling logic
                     if tool_calls and execute_tool_fn:
                         # Convert tool_calls to a serializable format for all providers
-                        serializable_tool_calls = []
-                        for tc in tool_calls:
-                            if isinstance(tc, dict):
-                                serializable_tool_calls.append(tc)  # Already a dict
-                            else:
-                                # Convert object to dict
-                                serializable_tool_calls.append({
-                                    "id": tc.id,
-                                    "type": getattr(tc, 'type', "function"),
-                                    "function": {
-                                        "name": tc.function.name,
-                                        "arguments": tc.function.arguments
-                                    }
-                                })
+                        serializable_tool_calls = self._serialize_tool_calls(tool_calls)
                         messages.append({
                             "role": "assistant",
                             "content": response_text,
@@ -826,20 +817,8 @@ class LLM:
                         tool_results = []  # Store all tool results
                         for tool_call in tool_calls:
                             # Handle both object and dict access patterns
-                            if isinstance(tool_call, dict):
-                                is_ollama = self._is_ollama_provider()
-                                function_name, arguments, tool_call_id = self._parse_tool_call_arguments(tool_call, is_ollama)
-                            else:
-                                # Handle object-style tool calls
-                                try:
-                                    function_name = tool_call.function.name
-                                    arguments = json.loads(tool_call.function.arguments) if tool_call.function.arguments else {}
-                                    tool_call_id = tool_call.id
-                                except (json.JSONDecodeError, AttributeError) as e:
-                                    logging.error(f"Error parsing object-style tool call: {e}")
-                                    function_name = "unknown_function"
-                                    arguments = {}
-                                    tool_call_id = f"tool_{id(tool_call)}"
+                            is_ollama = self._is_ollama_provider()
+                            function_name, arguments, tool_call_id = self._extract_tool_call_info(tool_call, is_ollama)
 
                             logging.debug(f"[TOOL_EXEC_DEBUG] About to execute tool {function_name} with args: {arguments}")
                             tool_result = execute_tool_fn(function_name, arguments)
@@ -1367,6 +1346,7 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                                 if delta.content:
                                     print("\033[K", end="\r")  
                                     print(f"Generating... {time.time() - start_time:.1f}s", end="\r")
+
                     else:
                         # Non-verbose streaming
                         async for chunk in await litellm.acompletion(
@@ -1380,9 +1360,12 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                         ):
                             if chunk and chunk.choices and chunk.choices[0].delta:
                                 delta = chunk.choices[0].delta
-                                response_text, tool_calls = self._process_stream_delta(
-                                    delta, response_text, tool_calls, formatted_tools
-                                )
+                                if delta.content:
+                                    response_text += delta.content
+                                
+                                # Capture tool calls from streaming chunks if provider supports it
+                                if formatted_tools and self._supports_streaming_tools():
+                                    tool_calls = self._process_tool_calls_from_stream(delta, tool_calls)
                     
                     response_text = response_text.strip()
                     
@@ -1417,20 +1400,7 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                 
                 if tool_calls:
                     # Convert tool_calls to a serializable format for all providers
-                    serializable_tool_calls = []
-                    for tc in tool_calls:
-                        if isinstance(tc, dict):
-                            serializable_tool_calls.append(tc)  # Already a dict
-                        else:
-                            # Convert object to dict
-                            serializable_tool_calls.append({
-                                "id": tc.id,
-                                "type": getattr(tc, 'type', "function"),
-                                "function": {
-                                    "name": tc.function.name,
-                                    "arguments": tc.function.arguments
-                                }
-                            })
+                    serializable_tool_calls = self._serialize_tool_calls(tool_calls)
                     messages.append({
                         "role": "assistant",
                         "content": response_text,
@@ -1440,20 +1410,8 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                     tool_results = []  # Store all tool results
                     for tool_call in tool_calls:
                         # Handle both object and dict access patterns
-                        if isinstance(tool_call, dict):
-                            is_ollama = self._is_ollama_provider()
-                            function_name, arguments, tool_call_id = self._parse_tool_call_arguments(tool_call, is_ollama)
-                        else:
-                            # Handle object-style tool calls
-                            try:
-                                function_name = tool_call.function.name
-                                arguments = json.loads(tool_call.function.arguments) if tool_call.function.arguments else {}
-                                tool_call_id = tool_call.id
-                            except (json.JSONDecodeError, AttributeError) as e:
-                                logging.error(f"Error parsing object-style tool call: {e}")
-                                function_name = "unknown_function"
-                                arguments = {}
-                                tool_call_id = f"tool_{id(tool_call)}"
+                        is_ollama = self._is_ollama_provider()
+                        function_name, arguments, tool_call_id = self._extract_tool_call_info(tool_call, is_ollama)
 
                         tool_result = await execute_tool_fn(function_name, arguments)
                         tool_results.append(tool_result)  # Store the result
@@ -1899,6 +1857,90 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
         
         return params
 
+    def _prepare_response_logging(self, temperature: float, stream: bool, verbose: bool, markdown: bool, **kwargs) -> Optional[Dict[str, Any]]:
+        """Prepare debug logging information for response methods"""
+        if logging.getLogger().getEffectiveLevel() == logging.DEBUG:
+            debug_info = {
+                "model": self.model,
+                "timeout": self.timeout,
+                "temperature": temperature,
+                "top_p": self.top_p,
+                "n": self.n,
+                "max_tokens": self.max_tokens,
+                "presence_penalty": self.presence_penalty,
+                "frequency_penalty": self.frequency_penalty,
+                "stream": stream,
+                "verbose": verbose,
+                "markdown": markdown,
+                "kwargs": str(kwargs)
+            }
+            return debug_info
+        return None
+
+    def _process_streaming_chunk(self, chunk) -> Optional[str]:
+        """Extract content from a streaming chunk"""
+        if chunk and chunk.choices and chunk.choices[0].delta.content:
+            return chunk.choices[0].delta.content
+        return None
+
+    def _process_tool_calls_from_stream(self, delta, tool_calls: List[Dict]) -> List[Dict]:
+        """Process tool calls from streaming delta chunks.
+        
+        This handles the accumulation of tool call data from streaming chunks,
+        building up the complete tool call information incrementally.
+        """
+        if hasattr(delta, 'tool_calls') and delta.tool_calls:
+            for tc in delta.tool_calls:
+                if tc.index >= len(tool_calls):
+                    tool_calls.append({
+                        "id": tc.id,
+                        "type": "function",
+                        "function": {"name": "", "arguments": ""}
+                    })
+                if tc.function.name:
+                    tool_calls[tc.index]["function"]["name"] = tc.function.name
+                if tc.function.arguments:
+                    tool_calls[tc.index]["function"]["arguments"] += tc.function.arguments
+        return tool_calls
+
+    def _serialize_tool_calls(self, tool_calls) -> List[Dict]:
+        """Convert tool calls to a serializable format for all providers."""
+        serializable_tool_calls = []
+        for tc in tool_calls:
+            if isinstance(tc, dict):
+                serializable_tool_calls.append(tc)  # Already a dict
+            else:
+                # Convert object to dict
+                serializable_tool_calls.append({
+                    "id": tc.id,
+                    "type": getattr(tc, 'type', "function"),
+                    "function": {
+                        "name": tc.function.name,
+                        "arguments": tc.function.arguments
+                    }
+                })
+        return serializable_tool_calls
+
+    def _extract_tool_call_info(self, tool_call, is_ollama: bool = False) -> tuple:
+        """Extract function name, arguments, and tool_call_id from a tool call.
+        
+        Handles both dict and object formats for tool calls.
+        """
+        if isinstance(tool_call, dict):
+            return self._parse_tool_call_arguments(tool_call, is_ollama)
+        else:
+            # Handle object-style tool calls
+            try:
+                function_name = tool_call.function.name
+                arguments = json.loads(tool_call.function.arguments) if tool_call.function.arguments else {}
+                tool_call_id = tool_call.id
+            except (json.JSONDecodeError, AttributeError) as e:
+                logging.error(f"Error parsing object-style tool call: {e}")
+                function_name = "unknown_function"
+                arguments = {}
+                tool_call_id = f"tool_{id(tool_call)}"
+            return function_name, arguments, tool_call_id
+
     # Response without tool calls
     def response(
         self,
@@ -1946,42 +1988,29 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
             )
 
             # Get response from LiteLLM
+            response_text = ""
+            completion_params = self._build_completion_params(
+                messages=messages,
+                temperature=temperature,
+                stream=stream,
+                **kwargs
+            )
+            
             if stream:
-                response_text = ""
                 if verbose:
                     with Live(display_generating("", start_time), console=console or self.console, refresh_per_second=4) as live:
-                        for chunk in litellm.completion(
-                            **self._build_completion_params(
-                                messages=messages,
-                                temperature=temperature,
-                                stream=True,
-                                **kwargs
-                            )
-                        ):
-                            if chunk and chunk.choices and chunk.choices[0].delta.content:
-                                content = chunk.choices[0].delta.content
+                        for chunk in litellm.completion(**completion_params):
+                            content = self._process_streaming_chunk(chunk)
+                            if content:
                                 response_text += content
                                 live.update(display_generating(response_text, start_time))
                 else:
-                    for chunk in litellm.completion(
-                        **self._build_completion_params(
-                            messages=messages,
-                            temperature=temperature,
-                            stream=True,
-                            **kwargs
-                        )
-                    ):
-                        if chunk and chunk.choices and chunk.choices[0].delta.content:
-                            response_text += chunk.choices[0].delta.content
+                    for chunk in litellm.completion(**completion_params):
+                        content = self._process_streaming_chunk(chunk)
+                        if content:
+                            response_text += content
             else:
-                response = litellm.completion(
-                    **self._build_completion_params(
-                        messages=messages,
-                        temperature=temperature,
-                        stream=False,
-                        **kwargs
-                    )
-                )
+                response = litellm.completion(**completion_params)
                 response_text = response.choices[0].message.content.strip()
 
             if verbose:
@@ -2022,6 +2051,7 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
             
             logger.debug("Using asynchronous response function")
             
+
             # Log all self values when in debug mode
             self._log_llm_config(
                 'Async response method',
@@ -2046,42 +2076,29 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
             )
 
             # Get response from LiteLLM
+            response_text = ""
+            completion_params = self._build_completion_params(
+                messages=messages,
+                temperature=temperature,
+                stream=stream,
+                **kwargs
+            )
+            
             if stream:
-                response_text = ""
                 if verbose:
                     with Live(display_generating("", start_time), console=console or self.console, refresh_per_second=4) as live:
-                        async for chunk in await litellm.acompletion(
-                            **self._build_completion_params(
-                                messages=messages,
-                                temperature=temperature,
-                                stream=True,
-                                **kwargs
-                            )
-                        ):
-                            if chunk and chunk.choices and chunk.choices[0].delta.content:
-                                content = chunk.choices[0].delta.content
+                        async for chunk in await litellm.acompletion(**completion_params):
+                            content = self._process_streaming_chunk(chunk)
+                            if content:
                                 response_text += content
                                 live.update(display_generating(response_text, start_time))
                 else:
-                    async for chunk in await litellm.acompletion(
-                        **self._build_completion_params(
-                            messages=messages,
-                            temperature=temperature,
-                            stream=True,
-                            **kwargs
-                        )
-                    ):
-                        if chunk and chunk.choices and chunk.choices[0].delta.content:
-                            response_text += chunk.choices[0].delta.content
+                    async for chunk in await litellm.acompletion(**completion_params):
+                        content = self._process_streaming_chunk(chunk)
+                        if content:
+                            response_text += content
             else:
-                response = await litellm.acompletion(
-                    **self._build_completion_params(
-                        messages=messages,
-                        temperature=temperature,
-                        stream=False,
-                        **kwargs
-                    )
-                )
+                response = await litellm.acompletion(**completion_params)
                 response_text = response.choices[0].message.content.strip()
 
             if verbose:

{praisonaiagents-0.0.120.dist-info → praisonaiagents-0.0.122.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: praisonaiagents
-Version: 0.0.120
+Version: 0.0.122
 Summary: Praison AI agents for completing complex tasks with Self Reflection Agents
 Author: Mervin Praison
 Requires-Python: >=3.10

{praisonaiagents-0.0.120.dist-info → praisonaiagents-0.0.122.dist-info}/RECORD

@@ -3,12 +3,12 @@ praisonaiagents/approval.py,sha256=UJ4OhfihpFGR5CAaMphqpSvqdZCHi5w2MGw1MByZ1FQ,9
 praisonaiagents/main.py,sha256=bamnEu5PaekloGi52VqAFclm-HzjEVeKtWF0Zpdmfzs,15479
 praisonaiagents/session.py,sha256=d-CZPYikOHb0q-H9f_IWKJsypnQfz1YKeLLkyxs6oDo,15532
 praisonaiagents/agent/__init__.py,sha256=IhIDtAkfJ99cxbttwou52coih_AejS2-jpazsX6LbDY,350
-praisonaiagents/agent/agent.py,sha256=_ROVyOTPBMB5Porv4YvZ4-kKWr4-tGMbSN7V8uDWZgk,109619
+praisonaiagents/agent/agent.py,sha256=BZx0iCP4hHfKFlfGlkZtanBJDDBDZe54dhf4Oskhmhw,110427
 praisonaiagents/agent/handoff.py,sha256=Saq0chqfvC6Zf5UbXvmctybbehqnotrXn72JsS-76Q0,13099
 praisonaiagents/agent/image_agent.py,sha256=-5MXG594HVwSpFMcidt16YBp7udtik-Cp7eXlzLE1fY,8696
 praisonaiagents/agents/__init__.py,sha256=_1d6Pqyk9EoBSo7E68sKyd1jDRlN1vxvVIRpoMc0Jcw,168
 praisonaiagents/agents/agents.py,sha256=WnptTEMSDMAM30Ka6rOAu6rBD-ZLev3qphb1a3BbP1g,63301
-praisonaiagents/agents/autoagents.py,sha256=NNSlqEsWf4up4lmdQwNl5_iTgodZ5aODUnjlXdp9vEQ,16127
+praisonaiagents/agents/autoagents.py,sha256=1stF8z94eyVg6hyfrLgdArlarftz_OFvEDtRMsGZFvg,21094
 praisonaiagents/guardrails/__init__.py,sha256=HA8zhp-KRHTxo0194MUwXOUJjPyjOu7E3d7xUIKYVVY,310
 praisonaiagents/guardrails/guardrail_result.py,sha256=2K1WIYRyT_s1H6vBGa-7HEHzXCFIyZXZVY4f0hnQyWc,1352
 praisonaiagents/guardrails/llm_guardrail.py,sha256=MTTqmYDdZX-18QN9T17T5P_6H2qnV8GVgymJufW1WuM,3277
@@ -16,7 +16,7 @@ praisonaiagents/knowledge/__init__.py,sha256=xL1Eh-a3xsHyIcU4foOWF-JdWYIYBALJH9b
 praisonaiagents/knowledge/chunking.py,sha256=G6wyHa7_8V0_7VpnrrUXbEmUmptlT16ISJYaxmkSgmU,7678
 praisonaiagents/knowledge/knowledge.py,sha256=OKPar-XGyAp1ndmbOOdCgqFnTCqpOThYVSIZRxZyP58,15683
 praisonaiagents/llm/__init__.py,sha256=6lTeQ8jWi1-KiwjCDCmkHo2e-bRLq2dP0s5iJWqjO3s,1421
-praisonaiagents/llm/llm.py,sha256=I08T3Du9PQndEzIEjDjacHqVkBpwg_AumcO4TsG85b8,107317
+praisonaiagents/llm/llm.py,sha256=8cDahPVMPI882J0psA1cXreJGXvO33eSOpMNy7FLCS4,107383
 praisonaiagents/llm/openai_client.py,sha256=0JvjCDHoH8I8kIt5vvObARkGdVaPWdTIv_FoEQ5EQPA,48973
 praisonaiagents/mcp/__init__.py,sha256=ibbqe3_7XB7VrIcUcetkZiUZS1fTVvyMy_AqCSFG8qc,240
 praisonaiagents/mcp/mcp.py,sha256=-fFx4MHffnN2woLnnV7Pzx3-1SFkn2j8Gp5F5ZIwKJ0,19698
@@ -53,7 +53,7 @@ praisonaiagents/tools/xml_tools.py,sha256=iYTMBEk5l3L3ryQ1fkUnNVYK-Nnua2Kx2S0dxN
 praisonaiagents/tools/yaml_tools.py,sha256=uogAZrhXV9O7xvspAtcTfpKSQYL2nlOTvCQXN94-G9A,14215
 praisonaiagents/tools/yfinance_tools.py,sha256=s2PBj_1v7oQnOobo2fDbQBACEHl61ftG4beG6Z979ZE,8529
 praisonaiagents/tools/train/data/generatecot.py,sha256=H6bNh-E2hqL5MW6kX3hqZ05g9ETKN2-kudSjiuU_SD8,19403
-praisonaiagents-0.0.120.dist-info/METADATA,sha256=0degpExWB64MNrEFvMEKEkopgjJp_UlALiDV8tFvxmk,1669
-praisonaiagents-0.0.120.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-praisonaiagents-0.0.120.dist-info/top_level.txt,sha256=_HsRddrJ23iDx5TTqVUVvXG2HeHBL5voshncAMDGjtA,16
-praisonaiagents-0.0.120.dist-info/RECORD,,
+praisonaiagents-0.0.122.dist-info/METADATA,sha256=pU6W0akH1O1raC15FOsFQW3GXuflwpznV2ij10vYUP4,1669
+praisonaiagents-0.0.122.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+praisonaiagents-0.0.122.dist-info/top_level.txt,sha256=_HsRddrJ23iDx5TTqVUVvXG2HeHBL5voshncAMDGjtA,16
+praisonaiagents-0.0.122.dist-info/RECORD,,