quantalogic 0.50.26__tar.gz → 0.50.29__tar.gz

{quantalogic-0.50.26 → quantalogic-0.50.29}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: quantalogic
-Version: 0.50.26
+Version: 0.50.29
 Summary: QuantaLogic ReAct Agents
 Author: Raphaël MANSUY
 Author-email: raphael.mansuy@gmail.com

{quantalogic-0.50.26 → quantalogic-0.50.29}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "quantalogic"
-version = "0.50.26"
+version = "0.50.29"
 description = "QuantaLogic ReAct Agents"
 authors = ["Raphaël MANSUY <raphael.mansuy@gmail.com>"]
 readme = "README.md"

{quantalogic-0.50.26 → quantalogic-0.50.29}/quantalogic/agent.py

@@ -1,10 +1,13 @@
 """Enhanced QuantaLogic agent implementing the ReAct framework."""
 
 import asyncio
+import os
 from collections.abc import Callable
 from datetime import datetime
+from pathlib import Path
 from typing import Any
 
+from jinja2 import Environment, FileSystemLoader
 from loguru import logger
 from pydantic import BaseModel, ConfigDict, PrivateAttr
 
@@ -559,17 +562,7 @@ class Agent(BaseModel):
         Returns:
             Generated summary text
         """
-        prompt_summary = (
-            "Summarize the conversation concisely:\n"
-            "format in markdown:\n"
-            "<thinking>\n"
-            " - 1. **Completed Steps**: Briefly describe the steps.\n"
-            " - 2. **Variables Used**: List the variables.\n"
-            " - 3. **Progress Analysis**: Assess progress.\n"
-            "</thinking>\n"
-            "Keep the summary clear and actionable.\n"
-        )
-
+        # Format conversation history for the template
         memory_copy = self.memory.memory.copy()
 
         if len(memory_copy) < 3:
@@ -578,6 +571,14 @@ class Agent(BaseModel):
 
         user_message = memory_copy.pop()
         assistant_message = memory_copy.pop()
+        
+        # Create summarization prompt using template
+        prompt_summary = self._render_template('memory_compaction_prompt.j2', 
+                                             conversation_history="\n\n".join(
+                                                 f"[{msg.role.upper()}]: {msg.content}" 
+                                                 for msg in memory_copy
+                                             ))
+        
         summary = await self.model.async_generate_with_history(messages_history=memory_copy, prompt=prompt_summary)
         
         # Remove last system message if present
@@ -620,15 +621,7 @@ class Agent(BaseModel):
             if len(content) < 1024 * 4:
                 return content
                 
-            prompt = (
-                "Create a task summary that captures ONLY: \n"
-                "1. Primary objective/purpose\n"
-                "2. Core actions/requirements\n"
-                "3. Desired end-state/outcome\n\n"
-                "Guidelines:\n"
-                "- Use imperative voice\n"
-                f"Input Task Description:\n{content}\n\n"
-            )
+            prompt = self._render_template('task_summary_prompt.j2', content=content)
             result = await self.model.async_generate(prompt=prompt)
             logger.debug(f"Generated summary: {result.response}")
             return result.response.strip() + "\n🚨 The FULL task is in <task> tag in the previous messages.\n"
@@ -785,16 +778,11 @@ class Agent(BaseModel):
             Tuple of (executed_tool_name, error_message)
         """
         repeat_count = self.last_tool_call.get("count", 0)
-        error_message = (
-            "Error: Detected repeated identical tool call pattern.\n"
-            f"Tool: {tool_name}\n"
-            f"Arguments: {arguments_with_values}\n"
-            f"Repeated {repeat_count} times\n\n"
-            "PLEASE:\n"
-            "1. Review your previous steps\n"
-            "2. Consider a different approach\n"
-            "3. Use a different tool or modify the arguments\n"
-            "4. Ensure you're making progress towards the goal"
+        error_message = self._render_template(
+            'repeated_tool_call_error.j2',
+            tool_name=tool_name,
+            arguments_with_values=arguments_with_values,
+            repeat_count=repeat_count
         )
         return tool_name, error_message
 
@@ -850,46 +838,19 @@ class Agent(BaseModel):
                 f"... content was truncated full content available by interpolation in variable {variable_name}"
             )
 
-        formatted_response = (
-            "# Analysis and Next Action Decision Point\n\n"
-            f"📊 Progress: Iteration {iteration}/{self.max_iterations}\n\n"
-            "## Global Task summary:\n"
-            f"```\n\n{self.task_to_solve_summary}```\n\n"
-            "## Available Resources\n"
-            f"🛠️ Tools:\n{self._get_tools_names_prompt()}\n\n"
-            f"📦 Variables:\n{self._get_variable_prompt()}\n\n"
-            "## Your Task\n"
-            "1. Analyze the execution result and progress, formalize if the current step is solved according to the task.\n"
-            "2. Determine the most effective next step\n"
-            "3. Select exactly ONE tool from the available list\n"
-            "4. Utilize variable interpolation where needed\n"
-            "## Response Requirements\n"
-            "Provide TWO markdown-formatted XML blocks:\n"
-            "1. Your analysis of the progression resulting from the execution of the tool in <thinking> tags, don't include <context_analysis/>\n"
-            "2. Your tool execution plan in <tool_name> tags\n\n"
-            "## Last executed action result\n"
-            f"Last executed tool {last_executed_tool} Execution Result:\n"
-            f"\n<{variable_name}>\n{response_display}\n</{variable_name}>\n"
-            "## Response Format\n"
-            "```xml\n"
-            "<thinking>\n"
-            "[Detailed analysis of progress, and reasoning for next step]\n"
-            "</thinking>\n"
-            "```\n"
-            "```xml\n"
-            "<action>\n"
-            "<selected_tool_name>\n"
-            "[Precise instruction for tool execution]\n"
-            "</selected_tool_name>\n"
-            "</action>\n"
-            "```\n\n"
-            "⚠️ Important:\n"
-            "- Respond ONLY with the two XML blocks\n"
-            "- No additional commentary\n"
-            "- If previous step failed, revise approach\n"
-            "- Use interpolated variables ($variable_name$) where required in tool calls, to minimize token usage, if possible\n"
-            "- strictly follow the required arguments for each tool as defined in system prompt\n"
-            "- Utilize <action><task_complete><answer>...</answer></task_complete><action> to indicate task completion, display the result or if the task is deemed unfeasible."
+        tools_prompt = self._get_tools_names_prompt()
+        variables_prompt = self._get_variable_prompt()
+        
+        formatted_response = self._render_template(
+            'observation_response_format.j2',
+            iteration=iteration,
+            max_iterations=self.max_iterations,
+            task_to_solve_summary=self.task_to_solve_summary,
+            tools_prompt=tools_prompt,
+            variables_prompt=variables_prompt,
+            last_executed_tool=last_executed_tool,
+            variable_name=variable_name,
+            response_display=response_display
         )
 
         return formatted_response
@@ -903,16 +864,14 @@ class Agent(BaseModel):
         Returns:
             The formatted task prompt
         """
-        prompt_task: str = (
-            "## Your task to solve:\n"
-            f"<task>\n{task}\n</task>\n"
-            "\n### Tools:\n"
-            "-----------------------------\n"
-            f"{self._get_tools_names_prompt()}\n"
-            "\n"
-            "### Variables:\n"
-            "-----------------------------\n"
-            f"{self._get_variable_prompt()}\n"
+        tools_prompt = self._get_tools_names_prompt()
+        variables_prompt = self._get_variable_prompt()
+        
+        prompt_task = self._render_template(
+            'task_prompt.j2',
+            task=task,
+            tools_prompt=tools_prompt,
+            variables_prompt=variables_prompt
         )
         return prompt_task
 
@@ -922,21 +881,8 @@ class Agent(BaseModel):
         Returns:
             Formatted tools prompt
         """
-        prompt_use_tools: str = (
-            "To accomplish this task, you have access to these tools:\n"
-            "\n"
-            f"{', '.join(self.tools.tool_names())}\n\n"
-            "Instructions:\n"
-            "\n"
-            "1. Select ONE tool per message\n"
-            "2. You will receive the tool's output in the next user response\n"
-            "3. Choose the most appropriate tool for each step\n"
-            "4. If it's not asked to write on files, don't use write_file tool\n"
-            "5. If files are written, then use tool to display the prepared download link\n"
-            "6. Give the final full answer using all the variables\n"
-            "7. Use task_complete tool to confirm task completion with the full content of the final answer\n"
-        )
-        return prompt_use_tools
+        tool_names = ', '.join(self.tools.tool_names())
+        return self._render_template('tools_prompt.j2', tool_names=tool_names)
 
     def _get_variable_prompt(self) -> str:
         """Construct a prompt that explains how to use variables.
@@ -944,17 +890,8 @@ class Agent(BaseModel):
         Returns:
             Formatted variables prompt
         """
-        prompt_use_variables: str = (
-            "To use a variable interpolation, use the format $variable_name$ in function arguments.\n"
-            "Example: <write_file><file_path>/path/to/file.txt</file_path><content>$var1$</content></write_file>\n"
-            "\n"
-            "Available variables:\n"
-            "\n"
-            f"{', '.join(self.variable_store.keys())}\n"
-            if len(self.variable_store.keys()) > 0
-            else "None\n"
-        )
-        return prompt_use_variables
+        variable_names = ', '.join(self.variable_store.keys()) if len(self.variable_store.keys()) > 0 else "None"
+        return self._render_template('variables_prompt.j2', variable_names=variable_names)
 
     def _calculate_context_occupancy(self) -> float:
         """Calculate the number of tokens in percentages for prompt and completion.
@@ -1063,4 +1000,31 @@ class Agent(BaseModel):
             tools_markdown=self.tools.to_markdown(),
             system_prompt=self.config.system_prompt,
         )
-        logger.debug(f"Set {len(tools)} tools")
+        logger.debug(f"Set {len(tools)} tools")
+        
+    def _render_template(self, template_name: str, **kwargs) -> str:
+        """Render a Jinja2 template with the provided variables.
+        
+        Args:
+            template_name: Name of the template file (without directory path)
+            **kwargs: Variables to pass to the template
+            
+        Returns:
+            str: The rendered template
+        """
+        try:
+            # Get the directory where this file is located
+            current_dir = Path(os.path.dirname(os.path.abspath(__file__)))
+            
+            # Set up Jinja2 environment
+            template_dir = current_dir / 'prompts'
+            env = Environment(loader=FileSystemLoader(template_dir))
+            
+            # Load the template
+            template = env.get_template(template_name)
+            
+            # Render the template with the provided variables
+            return template.render(**kwargs)
+        except Exception as e:
+            logger.error(f"Error rendering template {template_name}: {str(e)}")
+            raise

{quantalogic-0.50.26 → quantalogic-0.50.29}/quantalogic/model_info_litellm.py

@@ -1,7 +1,19 @@
 import functools
 
-import litellm
+# litellm will be imported lazily when needed
+_litellm = None
 
+def _get_litellm():
+    """Lazy load litellm module.
+
+    Returns:
+        The litellm module
+    """
+    global _litellm
+    if _litellm is None:
+        import litellm
+        _litellm = litellm
+    return _litellm
 
 @functools.lru_cache(maxsize=32)
 def litellm_get_model_info(model_name: str) -> dict | None:
@@ -16,6 +28,7 @@ def litellm_get_model_info(model_name: str) -> dict | None:
     Raises:
         ValueError: If model info cannot be found after prefix fallbacks
     """
+    litellm = _get_litellm()
     tried_models = [model_name]
 
     while True:
@@ -37,7 +50,6 @@ def litellm_get_model_info(model_name: str) -> dict | None:
 
     return None
 
-
 def litellm_get_model_max_input_tokens(model_name: str) -> int | None:
     """Get maximum input tokens for a model using litellm.
 
@@ -53,7 +65,6 @@ def litellm_get_model_max_input_tokens(model_name: str) -> int | None:
     except Exception:
         return 8192  # Default for many modern models
 
-
 def litellm_get_model_max_output_tokens(model_name: str) -> int | None:
     """Get maximum output tokens for a model using litellm.
 
@@ -67,4 +78,4 @@ def litellm_get_model_max_output_tokens(model_name: str) -> int | None:
         info = litellm_get_model_info(model_name)
         return info.get("max_output_tokens", 4096)
     except Exception:
-        return 4096  # Conservative default
+        return 4096  # Conservative default

quantalogic-0.50.29/quantalogic/prompts/memory_compaction_prompt.j2

@@ -0,0 +1,16 @@
+You are an AI assistant tasked with summarizing a conversation history.
+
+Create a concise summary that captures the essential information from the conversation, focusing on:
+1. The main task or problem being solved
+2. Key information discovered or shared
+3. Important decisions made
+4. Current status or progress
+
+Guidelines:
+- Be factual and objective
+- Preserve critical details and context
+- Focus on information that would be needed to continue the task
+- Aim for a summary that is about 20% of the original length
+
+Conversation History:
+{{ conversation_history }}

quantalogic-0.50.29/quantalogic/prompts/observation_response_format.j2

@@ -0,0 +1,55 @@
+# Analysis and Next Action Decision Point
+
+📊 Progress: Iteration {{ iteration }}/{{ max_iterations }}
+
+## Global Task summary:
+```
+
+{{ task_to_solve_summary }}```
+
+## Available Resources
+🛠️ Tools:
+{{ tools_prompt }}
+
+📦 Variables:
+{{ variables_prompt }}
+
+## Your Task
+1. Analyze the execution result and progress, formalize if the current step is solved according to the task.
+2. Determine the most effective next step
+3. Select exactly ONE tool from the available list
+4. Utilize variable interpolation where needed
+
+## Response Requirements
+Provide TWO markdown-formatted XML blocks:
+1. Your analysis of the progression resulting from the execution of the tool in <thinking> tags, don't include <context_analysis/>
+2. Your tool execution plan in <tool_name> tags
+
+## Last executed action result
+Last executed tool {{ last_executed_tool }} Execution Result:
+
+<{{ variable_name }}>
+{{ response_display }}
+</{{ variable_name }}>
+
+## Response Format
+```xml
+<thinking>
+[Detailed analysis of progress, and reasoning for next step]
+</thinking>
+```
+```xml
+<action>
+<selected_tool_name>
+[Precise instruction for tool execution]
+</selected_tool_name>
+</action>
+```
+
+⚠️ Important:
+- Respond ONLY with the two XML blocks
+- No additional commentary
+- If previous step failed, revise approach
+- Use interpolated variables ($variable_name$) where required in tool calls, to minimize token usage, if possible
+- strictly follow the required arguments for each tool as defined in system prompt
+- Utilize <action><task_complete><answer>...</answer></task_complete><action> to indicate task completion, display the result or if the task is deemed unfeasible.

quantalogic-0.50.29/quantalogic/prompts/repeated_tool_call_error.j2

@@ -0,0 +1,10 @@
+Error: Detected repeated identical tool call pattern.
+Tool: {{ tool_name }}
+Arguments: {{ arguments_with_values }}
+Repeated {{ repeat_count }} times
+
+PLEASE:
+1. Review your previous steps
+2. Consider a different approach
+3. Use a different tool or modify the arguments
+4. Ensure you're making progress towards the goal

quantalogic-0.50.26/quantalogic/prompts.py → quantalogic-0.50.29/quantalogic/prompts/system_prompt.j2

@@ -1,14 +1,8 @@
-from quantalogic.version import get_version
-
-
-def system_prompt(tools: str, environment: str, expertise: str = ""):
-    """System prompt for the ReAct chatbot with enhanced cognitive architecture."""
-    return f"""
-### Agent Identity: QuantaLogic {get_version()}
+### Agent Identity: QuantaLogic {{ version }}
 Expert ReAct AI Agent implementing OODA (Observe-Orient-Decide-Act) loop with advanced problem-solving capabilities.
 
 ### Domain Expertise
-{expertise}
+{{ expertise }}
 
 ### Input Protocol
 Task Format: <task>task_description</task>
@@ -76,14 +70,14 @@ Task Format: <task>task_description</task>
 ```xml
 <action>
 <task_complete>
-  <result>$final_output$</result>
+  <r>$final_output$</r>
 </task_complete>
 </action>
 ```
 
 ### Operational Parameters
-🛠️ Tools: {tools}
-🌐 Environment: {environment}
+🛠️ Tools: {{ tools }}
+🌐 Environment: {{ environment }}
 
 ### Execution Guidelines
 1. 🎯 Maintain laser focus on task objectives
@@ -94,5 +88,3 @@ Task Format: <task>task_description</task>
 6. 🛑 Fail fast and adapt when encountering blockers
 7. ✅ Verify completion criteria rigorously
 8. ✅ Return complete, full and usable results.
-
-"""

quantalogic-0.50.29/quantalogic/prompts/task_prompt.j2

@@ -0,0 +1,12 @@
+## Your task to solve:
+<task>
+{{ task }}
+</task>
+
+### Tools:
+-----------------------------
+{{ tools_prompt }}
+
+### Variables:
+-----------------------------
+{{ variables_prompt }}

quantalogic-0.50.29/quantalogic/prompts/task_summary_prompt.j2

@@ -0,0 +1,10 @@
+Create a task summary that captures ONLY: 
+1. Primary objective/purpose
+2. Core actions/requirements
+3. Desired end-state/outcome
+
+Guidelines:
+- Use imperative voice
+
+Input Task Description:
+{{ content }}

quantalogic-0.50.29/quantalogic/prompts/tools_prompt.j2

@@ -0,0 +1,13 @@
+To accomplish this task, you have access to these tools:
+
+{{ tool_names }}
+
+Instructions:
+
+1. Select ONE tool per message
+2. You will receive the tool's output in the next user response
+3. Choose the most appropriate tool for each step
+4. If it's not asked to write on files, don't use write_file tool
+5. If files are written, then use tool to display the prepared download link
+6. Give the final full answer using all the variables
+7. Use task_complete tool to confirm task completion with the full content of the final answer

quantalogic-0.50.29/quantalogic/prompts/variables_prompt.j2

@@ -0,0 +1,6 @@
+To use a variable interpolation, use the format $variable_name$ in function arguments.
+Example: <write_file><file_path>/path/to/file.txt</file_path><content>$var1$</content></write_file>
+
+Available variables:
+
+{{ variable_names }}

quantalogic-0.50.29/quantalogic/prompts.py

@@ -0,0 +1,38 @@
+import os
+from pathlib import Path
+
+from jinja2 import Environment, FileSystemLoader
+
+from quantalogic.version import get_version
+
+
+def system_prompt(tools: str, environment: str, expertise: str = ""):
+    """System prompt for the ReAct chatbot with enhanced cognitive architecture.
+    
+    Uses a Jinja2 template from the prompts directory.
+    
+    Args:
+        tools: Available tools for the agent
+        environment: Environment information
+        expertise: Domain expertise information
+        
+    Returns:
+        str: The rendered system prompt
+    """
+    # Get the directory where this file is located
+    current_dir = Path(os.path.dirname(os.path.abspath(__file__)))
+    
+    # Set up Jinja2 environment
+    template_dir = current_dir / 'prompts'
+    env = Environment(loader=FileSystemLoader(template_dir))
+    
+    # Load the template
+    template = env.get_template('system_prompt.j2')
+    
+    # Render the template with the provided variables
+    return template.render(
+        version=get_version(),
+        tools=tools,
+        environment=environment,
+        expertise=expertise
+    )

{quantalogic-0.50.26 → quantalogic-0.50.29}/quantalogic/quantlitellm.py

@@ -38,7 +38,10 @@ PROVIDERS = {
         env_var="DASHSCOPE_API_KEY",
     ),
     "nvidia": ModelProviderConfig(
-        prefix="nvidia/", provider="openai", base_url="https://integrate.api.nvidia.com/v1", env_var="NVIDIA_API_KEY"
+        prefix="nvidia/",
+        provider="openai",
+        base_url="https://integrate.api.nvidia.com/v1",
+        env_var="NVIDIA_API_KEY",
     ),
     "ovh": ModelProviderConfig(
         prefix="ovh/",
@@ -160,4 +163,4 @@ def get_model_max_output_tokens(model_name: str) -> int | None:
 
 def get_model_info(model_name: str) -> dict | None:
     """Get model information for a given model name."""
-    return model_info.get(model_name, None)
+    return model_info.get(model_name, None)