quantalogic 0.2.15__tar.gz → 0.2.17__tar.gz

{quantalogic-0.2.15 → quantalogic-0.2.17}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: quantalogic
-Version: 0.2.15
+Version: 0.2.17
 Summary: QuantaLogic ReAct Agents
 Author: Raphaël MANSUY
 Author-email: raphael.mansuy@gmail.com
@@ -15,11 +15,20 @@ Requires-Dist: fastapi (>=0.115.6,<0.116.0)
 Requires-Dist: google-auth (>=2.20.0,<3.0.0)
 Requires-Dist: google-search-results (>=2.4.2,<3.0.0)
 Requires-Dist: litellm (>=1.56.4,<2.0.0)
+Requires-Dist: llmlingua (>=0.2.2,<0.3.0)
 Requires-Dist: loguru (>=0.7.3,<0.8.0)
 Requires-Dist: markitdown (>=0.0.1a3,<0.0.2)
+Requires-Dist: mkdocs-git-revision-date-localized-plugin (>=1.2.0,<2.0.0)
+Requires-Dist: mkdocs-macros-plugin (>=1.0.4,<2.0.0)
+Requires-Dist: mkdocs-material[imaging] (>=9.5.49,<10.0.0)
+Requires-Dist: mkdocs-mermaid2-plugin (>=1.1.1,<2.0.0)
+Requires-Dist: mkdocs-minify-plugin (>=0.7.1,<0.8.0)
+Requires-Dist: mkdocstrings (>=0.24.0,<0.25.0)
+Requires-Dist: mkdocstrings-python (>=1.7.0,<2.0.0)
 Requires-Dist: pathspec (>=0.12.1,<0.13.0)
 Requires-Dist: prompt-toolkit (>=3.0.48,<4.0.0)
 Requires-Dist: pydantic (>=2.10.4,<3.0.0)
+Requires-Dist: pymdown-extensions (>=10.3.1,<11.0.0)
 Requires-Dist: rich (>=13.9.4,<14.0.0)
 Requires-Dist: serpapi (>=0.1.5,<0.2.0)
 Requires-Dist: tenacity (>=9.0.0,<10.0.0)
@@ -43,7 +52,7 @@ Description-Content-Type: text/markdown
 
 [![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 [![Python](https://img.shields.io/badge/Python-3.12+-blue.svg)](https://www.python.org/downloads/)
-[![Documentation](https://img.shields.io/badge/docs-latest-brightgreen.svg)]()
+[![Documentation](https://img.shields.io/badge/docs-latest-brightgreen.svg)](https://quantalogic.github.io/quantalogic/)
 
 QuantaLogic is a  ReAct (Reasoning & Action) framework for building advanced AI agents. 
 
@@ -51,6 +60,8 @@ It seamlessly integrates large language models (LLMs) with a robust tool system,
 
 The `cli` version include coding capabilities comparable to Aider.
 
+[📖 Documentation](https://quantalogic.github.io/quantalogic/)
+
 
 ## Why QuantaLogic?
 
@@ -80,6 +91,7 @@ We created [QuantaLogic](https://www.quantalogic.app) because we saw a significa
 - [Development](#-development)
 - [Contributing](#-contributing)
 - [License](#-license)
+- [Documentation Development](#-documentation-development)
 
 ## 📦 Installation
 
@@ -128,12 +140,13 @@ Usage: quantalogic [OPTIONS] COMMAND [ARGS]...
 Options:
   --version                       Show version information.
   --model-name TEXT               Specify the text model to use (litellm format,
-                                  e.g. "openrouter/deepseek-chat").
+                                  e.g. "openrouter/deepseek/deepseek-chat").
   --vision-model-name TEXT        Specify the vision model to use (litellm format,
-                                  e.g. "openrouter/A/gpt-4o-mini").
+                                  e.g. "openrouter/openai/gpt-4o-mini").
   --log [info|debug|warning]      Set logging level (info/debug/warning).
   --verbose                       Enable verbose output.
-  --mode [code|basic|interpreter|full|code-basic|search]
+  --max-iterations INTEGER        Maximum iterations for task solving (default: 30).
+  --mode [code|basic|interpreter|full|code-basic|search|search-full]
                                   Agent mode (code/search/full).
   --help                          Show this message and exit.
 
@@ -154,6 +167,7 @@ task    Execute a task with the QuantaLogic AI Assistant
 - interpreter: Interactive code execution agent
 - full: Full-featured agent with all capabilities
 - code-basic: Coding agent with basic reasoning
+- search: Web search agent with Wikipedia, DuckDuckGo and SERPApi integration
 
 #### Task Execution
 
@@ -249,7 +263,7 @@ from quantalogic.tools import PythonTool, ReadFileTool
 
 # Create agent with specific tools
 agent = Agent(
-    model_name="openrouter/deepseek-chat",
+    model_name="openrouter/deepseek/deepseek-chat",
     tools=[
         PythonTool(),
         ReadFileTool()
@@ -842,148 +856,7 @@ print(results)
 ```
 ```
 
-#### Creating Custom Tools
-
-```python
-from quantalogic.tools import Tool, ToolArgument
-
-class DatabaseTool(Tool):
-    name: str = "database_tool"
-    description: str = "Execute database operations"
-    need_validation: bool = True
-    
-    arguments: list[ToolArgument] = [
-        ToolArgument(
-            name="query",
-            arg_type="string",
-            description="SQL query to execute",
-            required=True
-        )
-    ]
-    
-    def execute(self, query: str) -> str:
-        # Tool implementation
-        return "Query results"
 ```
-
-
-## 🌐 Web Interface
-
-Features:
-- Real-time event visualization
-- Task submission and monitoring
-- Interactive validation dialogs
-- Model selection
-- Event filtering and search
-
-### API Endpoints
-
-| Endpoint           | Method | Description     |
-| ------------------ | ------ | --------------- |
-| `/tasks`           | POST   | Submit tasks    |
-| `/tasks/{task_id}` | GET    | Task status     |
-| `/events`          | GET    | SSE endpoint    |
-| `/validate`        | POST   | Task validation |
-
-
-## 📖 Examples
-
-### Python Tool Integration Example
-
-```python
-import os
-
-from quantalogic import Agent, console_print_events
-from quantalogic.tools import (
-    PythonTool,
-)
-
-# Verify API key is set - required for authentication with DeepSeek's API
-# This check ensures the agent won't fail during runtime due to missing credentials
-if not os.environ.get("DEEPSEEK_API_KEY"):
-    raise ValueError("DEEPSEEK_API_KEY environment variable is not set")
-
-# Initialize agent with DeepSeek model and Python tool
-agent = Agent(model_name="deepseek/deepseek-chat", tools=[PythonTool()])
-
-# Configure comprehensive event monitoring system
-# Tracks all agent activities including:
-# - Code execution steps
-# - Tool interactions
-# - Error conditions
-# Essential for debugging and performance optimization
-agent.event_emitter.on(
-    "*",
-    console_print_events,
-)
-
-# Execute a precision mathematics task demonstrating:
-# - High-precision calculations
-# - PythonTool integration
-# - Real-time monitoring capabilities
-result = agent.solve_task("1. Calculate PI with 10000 decimal places.")
-print(result)
-```
-
-### Agent with Event Monitoring
-
-```python
-import os
-
-from quantalogic import Agent, console_print_events
-from quantalogic.tools import (
-    LLMTool,
-)
-
-# Verify API key is set - required for authentication with DeepSeek's API
-# This check ensures the agent won't fail during runtime due to missing credentials
-if not os.environ.get("DEEPSEEK_API_KEY"):
-    raise ValueError("DEEPSEEK_API_KEY environment variable is not set")
-
-# Initialize agent with DeepSeek model and LLM tool
-# The LLM tool serves dual purpose:
-# 1. As a reasoning engine for the agent's cognitive processes
-# 2. As a latent space explorer, enabling the agent to:
-#    - Discover novel solution paths
-#    - Generate creative combinations of concepts
-#    - Explore alternative reasoning strategies
-# Using the same model ensures consistent behavior across both roles
-agent = Agent(model_name="deepseek/deepseek-chat", tools=[LLMTool(model_name="deepseek/deepseek-chat")])
-
-# Set up event monitoring to track agent's lifecycle
-# This helps in debugging and understanding the agent's behavior
-agent.event_emitter.on(
-    [
-        "task_complete",
-        "task_think_start",
-        "task_think_end",
-        "tool_execution_start",
-        "tool_execution_end",
-        "error_max_iterations_reached",
-        "memory_full",
-        "memory_compacted",
-        "memory_summary",
-    ],
-    console_print_events,
-)
-
-# Execute a multi-step task showcasing agent's capabilities
-# Demonstrates:
-# 1. Creative content generation
-# 2. Language translation
-# 3. Style adaptation
-# 4. Multi-step reasoning and execution
-result = agent.solve_task(
-    "1. Write a poem in English about a dog. "
-    "2. Translate the poem into French. "
-    "3. Choose 2 French authors"
-    "4. Rewrite the translated poem with the style of the chosen authors. "
-)
-print(result)
-```
-
-
-
 ### Project Documentation
 
 ```python
@@ -991,7 +864,7 @@ from quantalogic import Agent
 from quantalogic.tools import MarkitdownTool, ReadFileTool
 
 agent = Agent(
-    model_name="openrouter/deepseek-chat",
+    model_name="openrouter/deepseek/deepseek-chat",
     tools=[MarkitdownTool(), ReadFileTool()]
 )
 

{quantalogic-0.2.15 → quantalogic-0.2.17}/README.md

@@ -2,7 +2,7 @@
 
 [![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 [![Python](https://img.shields.io/badge/Python-3.12+-blue.svg)](https://www.python.org/downloads/)
-[![Documentation](https://img.shields.io/badge/docs-latest-brightgreen.svg)]()
+[![Documentation](https://img.shields.io/badge/docs-latest-brightgreen.svg)](https://quantalogic.github.io/quantalogic/)
 
 QuantaLogic is a  ReAct (Reasoning & Action) framework for building advanced AI agents. 
 
@@ -10,6 +10,8 @@ It seamlessly integrates large language models (LLMs) with a robust tool system,
 
 The `cli` version include coding capabilities comparable to Aider.
 
+[📖 Documentation](https://quantalogic.github.io/quantalogic/)
+
 
 ## Why QuantaLogic?
 
@@ -39,6 +41,7 @@ We created [QuantaLogic](https://www.quantalogic.app) because we saw a significa
 - [Development](#-development)
 - [Contributing](#-contributing)
 - [License](#-license)
+- [Documentation Development](#-documentation-development)
 
 ## 📦 Installation
 
@@ -87,12 +90,13 @@ Usage: quantalogic [OPTIONS] COMMAND [ARGS]...
 Options:
   --version                       Show version information.
   --model-name TEXT               Specify the text model to use (litellm format,
-                                  e.g. "openrouter/deepseek-chat").
+                                  e.g. "openrouter/deepseek/deepseek-chat").
   --vision-model-name TEXT        Specify the vision model to use (litellm format,
-                                  e.g. "openrouter/A/gpt-4o-mini").
+                                  e.g. "openrouter/openai/gpt-4o-mini").
   --log [info|debug|warning]      Set logging level (info/debug/warning).
   --verbose                       Enable verbose output.
-  --mode [code|basic|interpreter|full|code-basic|search]
+  --max-iterations INTEGER        Maximum iterations for task solving (default: 30).
+  --mode [code|basic|interpreter|full|code-basic|search|search-full]
                                   Agent mode (code/search/full).
   --help                          Show this message and exit.
 
@@ -113,6 +117,7 @@ task    Execute a task with the QuantaLogic AI Assistant
 - interpreter: Interactive code execution agent
 - full: Full-featured agent with all capabilities
 - code-basic: Coding agent with basic reasoning
+- search: Web search agent with Wikipedia, DuckDuckGo and SERPApi integration
 
 #### Task Execution
 
@@ -208,7 +213,7 @@ from quantalogic.tools import PythonTool, ReadFileTool
 
 # Create agent with specific tools
 agent = Agent(
-    model_name="openrouter/deepseek-chat",
+    model_name="openrouter/deepseek/deepseek-chat",
     tools=[
         PythonTool(),
         ReadFileTool()
@@ -801,148 +806,7 @@ print(results)
 ```
 ```
 
-#### Creating Custom Tools
-
-```python
-from quantalogic.tools import Tool, ToolArgument
-
-class DatabaseTool(Tool):
-    name: str = "database_tool"
-    description: str = "Execute database operations"
-    need_validation: bool = True
-    
-    arguments: list[ToolArgument] = [
-        ToolArgument(
-            name="query",
-            arg_type="string",
-            description="SQL query to execute",
-            required=True
-        )
-    ]
-    
-    def execute(self, query: str) -> str:
-        # Tool implementation
-        return "Query results"
 ```
-
-
-## 🌐 Web Interface
-
-Features:
-- Real-time event visualization
-- Task submission and monitoring
-- Interactive validation dialogs
-- Model selection
-- Event filtering and search
-
-### API Endpoints
-
-| Endpoint           | Method | Description     |
-| ------------------ | ------ | --------------- |
-| `/tasks`           | POST   | Submit tasks    |
-| `/tasks/{task_id}` | GET    | Task status     |
-| `/events`          | GET    | SSE endpoint    |
-| `/validate`        | POST   | Task validation |
-
-
-## 📖 Examples
-
-### Python Tool Integration Example
-
-```python
-import os
-
-from quantalogic import Agent, console_print_events
-from quantalogic.tools import (
-    PythonTool,
-)
-
-# Verify API key is set - required for authentication with DeepSeek's API
-# This check ensures the agent won't fail during runtime due to missing credentials
-if not os.environ.get("DEEPSEEK_API_KEY"):
-    raise ValueError("DEEPSEEK_API_KEY environment variable is not set")
-
-# Initialize agent with DeepSeek model and Python tool
-agent = Agent(model_name="deepseek/deepseek-chat", tools=[PythonTool()])
-
-# Configure comprehensive event monitoring system
-# Tracks all agent activities including:
-# - Code execution steps
-# - Tool interactions
-# - Error conditions
-# Essential for debugging and performance optimization
-agent.event_emitter.on(
-    "*",
-    console_print_events,
-)
-
-# Execute a precision mathematics task demonstrating:
-# - High-precision calculations
-# - PythonTool integration
-# - Real-time monitoring capabilities
-result = agent.solve_task("1. Calculate PI with 10000 decimal places.")
-print(result)
-```
-
-### Agent with Event Monitoring
-
-```python
-import os
-
-from quantalogic import Agent, console_print_events
-from quantalogic.tools import (
-    LLMTool,
-)
-
-# Verify API key is set - required for authentication with DeepSeek's API
-# This check ensures the agent won't fail during runtime due to missing credentials
-if not os.environ.get("DEEPSEEK_API_KEY"):
-    raise ValueError("DEEPSEEK_API_KEY environment variable is not set")
-
-# Initialize agent with DeepSeek model and LLM tool
-# The LLM tool serves dual purpose:
-# 1. As a reasoning engine for the agent's cognitive processes
-# 2. As a latent space explorer, enabling the agent to:
-#    - Discover novel solution paths
-#    - Generate creative combinations of concepts
-#    - Explore alternative reasoning strategies
-# Using the same model ensures consistent behavior across both roles
-agent = Agent(model_name="deepseek/deepseek-chat", tools=[LLMTool(model_name="deepseek/deepseek-chat")])
-
-# Set up event monitoring to track agent's lifecycle
-# This helps in debugging and understanding the agent's behavior
-agent.event_emitter.on(
-    [
-        "task_complete",
-        "task_think_start",
-        "task_think_end",
-        "tool_execution_start",
-        "tool_execution_end",
-        "error_max_iterations_reached",
-        "memory_full",
-        "memory_compacted",
-        "memory_summary",
-    ],
-    console_print_events,
-)
-
-# Execute a multi-step task showcasing agent's capabilities
-# Demonstrates:
-# 1. Creative content generation
-# 2. Language translation
-# 3. Style adaptation
-# 4. Multi-step reasoning and execution
-result = agent.solve_task(
-    "1. Write a poem in English about a dog. "
-    "2. Translate the poem into French. "
-    "3. Choose 2 French authors"
-    "4. Rewrite the translated poem with the style of the chosen authors. "
-)
-print(result)
-```
-
-
-
 ### Project Documentation
 
 ```python
@@ -950,7 +814,7 @@ from quantalogic import Agent
 from quantalogic.tools import MarkitdownTool, ReadFileTool
 
 agent = Agent(
-    model_name="openrouter/deepseek-chat",
+    model_name="openrouter/deepseek/deepseek-chat",
     tools=[MarkitdownTool(), ReadFileTool()]
 )
 

{quantalogic-0.2.15 → quantalogic-0.2.17}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "quantalogic"
-version = "0.2.15"
+version = "0.2.17"
 description = "QuantaLogic ReAct Agents"
 authors = ["Raphaël MANSUY <raphael.mansuy@gmail.com>"]
 readme = "README.md"
@@ -37,9 +37,21 @@ types-requests = "^2.32.0.20241016"
 google-search-results = "^2.4.2"
 serpapi = "^0.1.5"
 duckduckgo-search = "^7.2.1"
+mkdocs-material = {extras = ["imaging"], version = "^9.5.49"}
+mkdocs-mermaid2-plugin = "^1.1.1"
+mkdocs-minify-plugin = "^0.7.1"
+mkdocs-git-revision-date-localized-plugin = "^1.2.0"
+mkdocs-macros-plugin = "^1.0.4"
+mkdocstrings = "^0.24.0"
+mkdocstrings-python = "^1.7.0"
+pymdown-extensions = "^10.3.1"
+llmlingua = "^0.2.2"
 
 [tool.poetry.scripts]
 quantalogic = "quantalogic.main:cli"
+docs-serve = "quantalogic.docs_cli:serve_docs"
+docs-build = "quantalogic.docs_cli:build_docs"
+docs-deploy = "quantalogic.docs_cli:deploy_docs"
 
 [tool.poetry.group.dev.dependencies]
 ruff = "^0.8.4"
@@ -47,6 +59,17 @@ pytest = "^8.2.0"
 pytest-mock = "^3.14.0"
 litellm = "^1.55.9"
 ollama = "^0.4.4"
+mkdocs = "^1.6.0"
+mkdocs-material = "^9.5.0"
+
+[tool.poetry.group.docs.dependencies]
+mkdocs-material = {extras = ["imaging"], version = "^9.5.49"}
+mkdocs-mermaid2-plugin = "^1.1.1"
+mkdocs-git-revision-date-localized-plugin = "^1.2.0"
+mkdocs-macros-plugin = "^1.0.4"
+mkdocstrings = "^0.24.0"
+mkdocstrings-python = "^1.7.0"
+pymdown-extensions = "^10.3.1"
 
 [build-system]
 requires = ["poetry-core"]

{quantalogic-0.2.15 → quantalogic-0.2.17}/quantalogic/__init__.py

@@ -11,10 +11,11 @@ warnings.filterwarnings(
 
 
 from .agent import Agent  # noqa: E402
+from .console_print_events import console_print_events  # noqa: E402
+from .console_print_token import console_print_token  # noqa: E402
 from .event_emitter import EventEmitter  # noqa: E402
 from .memory import AgentMemory, VariableMemory  # noqa: E402
-from .print_event import console_print_events  # noqa: E402
 
 """QuantaLogic package for AI-powered generative models."""
 
-__all__ = ["Agent", "EventEmitter", "AgentMemory", "VariableMemory", "console_print_events"]
+__all__ = ["Agent", "EventEmitter", "AgentMemory", "VariableMemory", "console_print_events","console_print_token"]

{quantalogic-0.2.15 → quantalogic-0.2.17}/quantalogic/agent.py

@@ -8,7 +8,7 @@ from loguru import logger
 from pydantic import BaseModel, ConfigDict
 
 from quantalogic.event_emitter import EventEmitter
-from quantalogic.generative_model import GenerativeModel
+from quantalogic.generative_model import GenerativeModel, ResponseStats, TokenUsage
 from quantalogic.memory import AgentMemory, Message, VariableMemory
 from quantalogic.prompts import system_prompt
 from quantalogic.tool_manager import ToolManager
@@ -85,6 +85,9 @@ class Agent(BaseModel):
         """Initialize the agent with model, memory, tools, and configurations."""
         try:
             logger.debug("Initializing agent...")
+            # Create event emitter first
+            event_emitter = EventEmitter()
+
             # Add TaskCompleteTool to the tools list if not already present
             if TaskCompleteTool() not in tools:
                 tools.append(TaskCompleteTool())
@@ -108,7 +111,7 @@ class Agent(BaseModel):
 
             logger.debug("Base class init started ...")
             super().__init__(
-                model=GenerativeModel(model=model_name),
+                model=GenerativeModel(model=model_name, event_emitter=event_emitter),
                 memory=memory,
                 variable_store=VariableMemory(),
                 tools=tool_manager,
@@ -116,19 +119,21 @@ class Agent(BaseModel):
                 ask_for_user_validation=ask_for_user_validation,
                 task_to_solve=task_to_solve,
                 specific_expertise=specific_expertise,
+                event_emitter=event_emitter,
             )
             logger.debug("Agent initialized successfully.")
         except Exception as e:
             logger.error(f"Failed to initialize agent: {str(e)}")
             raise
 
-    def solve_task(self, task: str, max_iterations: int = 30) -> str:
+    def solve_task(self, task: str, max_iterations: int = 30, streaming: bool = False) -> str:
         """Solve the given task using the ReAct framework.
 
         Args:
             task (str): The task description.
             max_iterations (int, optional): Maximum number of iterations to attempt solving the task.
                 Defaults to 30 to prevent infinite loops and ensure timely task completion.
+            streaming (bool, optional): Whether to use streaming mode for generating responses.
 
         Returns:
             str: The final response after task completion.
@@ -172,11 +177,34 @@ class Agent(BaseModel):
 
                 self._compact_memory_if_needed(current_prompt)
 
-                result = self.model.generate_with_history(messages_history=self.memory.memory, prompt=current_prompt)
+                if streaming:
+                    # For streaming, collect the response chunks
+                    content = ""
+                    for chunk in self.model.generate_with_history(
+                        messages_history=self.memory.memory, prompt=current_prompt, streaming=True
+                    ):
+                        content += chunk
+
+                    # Create a response object similar to non-streaming mode
+                    result = ResponseStats(
+                        response=content,
+                        usage=TokenUsage(
+                            prompt_tokens=0,  # We don't have token counts in streaming mode
+                            completion_tokens=0,
+                            total_tokens=0,
+                        ),
+                        model=self.model.model,
+                        finish_reason="stop",
+                    )
+                else:
+                    result = self.model.generate_with_history(
+                        messages_history=self.memory.memory, prompt=current_prompt, streaming=False
+                    )
 
                 content = result.response
-                token_usage = result.usage
-                self.total_tokens = token_usage.total_tokens
+                if not streaming:  # Only update tokens for non-streaming mode
+                    token_usage = result.usage
+                    self.total_tokens = token_usage.total_tokens
 
                 # Emit event: Task Think End
                 self._emit_event(
@@ -187,7 +215,7 @@ class Agent(BaseModel):
                 )
 
                 # Process the assistant's response
-                result = self._observe_response(result.response, iteration=self.current_iteration)
+                result = self._observe_response(content, iteration=self.current_iteration)
 
                 current_prompt = result.next_prompt
 
@@ -292,9 +320,10 @@ class Agent(BaseModel):
                 is_repeated_call = self._is_repeated_tool_call(tool_name, arguments_with_values)
 
                 if is_repeated_call:
-                    return self._handle_repeated_tool_call(tool_name, arguments_with_values)
+                    executed_tool, response = self._handle_repeated_tool_call(tool_name, arguments_with_values)
+                else:
+                    executed_tool, response = self._execute_tool(tool_name, tool, arguments_with_values)
 
-                executed_tool, response = self._execute_tool(tool_name, tool, arguments_with_values)
                 if not executed_tool:
                     return self._handle_tool_execution_failure(response)
 
@@ -406,7 +435,7 @@ class Agent(BaseModel):
 
         formatted_response = (
             "\n"
-            f"--- Observations for iteration {iteration} ---\n"
+            f"--- Observations for iteration {iteration} / max {self.max_iterations} ---\n"
             "\n"
             f"\n --- Tool execution result stored in variable ${variable_name}$ --- \n"
             "\n"
@@ -427,34 +456,26 @@ class Agent(BaseModel):
 
         # Format the response message
         formatted_response = (
-            "\n"
-            f"--- Observations for iteration {iteration} ---\n"
-            "\n"
-            f"\n --- Tool execution result stored in variable ${variable_name}$ --- \n"
-            "\n"
-            f"<{variable_name}>\n{response_display}\n</{variable_name}>\n" + "\n"
-            "\n"
-            f"--- Tools --- \n"
-            "\n"
-            f"{self._get_tools_names_prompt()}"
-            "\n"
-            f"--- Variables --- \n"
-            "\n"
-            f"{self._get_variable_prompt()}"
-            "\n"
-            "You must analyze this answer and evaluate what to do next to solve the task.\n"
-            "If the step failed, take a step back and rethink your approach.\n"
-            "\n"
-            "--- Task to solve summary ---\n"
-            "\n"
-            f"{self.task_to_solve_summary}"
-            "\n"
+            f"\n--- Observations for iteration {iteration} / max {self.max_iterations} ---\n"
+            f"\n--- Tool execution result in ${variable_name}$ ---\n"
+            f"<{variable_name}>\n{response_display}\n</{variable_name}>\n\n"
+            f"--- Tools ---\n{self._get_tools_names_prompt()}\n"
+            f"--- Variables ---\n{self._get_variable_prompt()}\n"
+            "Analyze this response to determine the next steps. If the step failed, reconsider your approach.\n"
+            f"--- Task to solve summary ---\n{self.task_to_solve_summary}\n"
             "--- Format ---\n"
-            "\n"
-            "You MUST respond with exactly two XML blocks formatted in markdown:\n"
-            "\n"
-            " - One <thinking> block detailing your analysis,\n"
-            " - One <tool_name> block specifying the chosen tool and its arguments, as outlined in the system prompt.\n"
+            "Respond only with two XML blocks in markdown as specified in system prompt.\n"
+            "No extra comments must be added.\n"
+            "```xml\n"
+            "<thinking>\n"
+            "...\n"
+            "</thinking>\n"
+            "```\n"
+            "```xml\n"
+            "< ...tool_name... >\n"
+            "...\n"
+            "</ ...tool_name... >\n"
+            "```"
         )
 
         return formatted_response