built-from-scratch/ReAct-agent.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook goes through the process of building a ReAct agent from scratch. [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/synacktraa/ai-agents/blob/main/built-from-scratch/ReAct-agent.ipynb)\n",
    "\n",
    "**Tutorial by James Murdza**: \n",
    "<a href=\"https://youtu.be/C0QdSBoJiMs\">\n",
    "    <img src=\"https://img.shields.io/badge/YouTube-FF0000?style=for-the-badge&logo=youtube&logoColor=white\" alt=\"YouTube\" width=90px height=20px>\n",
    "</a>\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture --no-stderr\n",
    "%pip install \"docstring-parser>=0.16\" duckduckgo-search litellm loguru requests sympy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implementing a `ReActAgent` compatible tool\n",
    "\n",
    "First, we're going to implement a `ReActTool` class. \n",
    "\n",
    "It will be responsible for:\n",
    "\n",
    "- Wrapping a function that must have exactly one parameter of a supported type (string, integer, float, boolean, list, dictionary, tuple, or set). If no type is specified for the parameter, it will be treated as a string. The function must also include a docstring that describes what it does.\n",
    "\n",
    "- Converting raw string inputs into the correct type before passing them to the wrapped function. For example, if the wrapped function expects an integer, it will automatically convert string inputs like \"123\" into the integer 123.\n",
    "\n",
    "- Catches all exception raised by the associated function and returns a `ExceptionInfo` instead of just halting the execution. The exception info can be used by agent to understand what went wrong and how to fix it. This also helps in when defining the function, as we don't have to explicitly catch and handle exceptions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import inspect\n",
    "from dataclasses import dataclass\n",
    "from typing import Any, Callable, Generic, Protocol, TypeVar\n",
    "\n",
    "from docstring_parser import parse\n",
    "\n",
    "\n",
    "T_Param = TypeVar(\"T_Param\", str, int, float, bool, list, dict, tuple, set)\n",
    "\"\"\"Type representing the accepted parameter types for the function.\"\"\"\n",
    "\n",
    "class SupportsStr(Protocol):\n",
    "    def __str__(self) -> str:\n",
    "        ...\n",
    "\n",
    "T_Retval = TypeVar(\"T_Retval\", bound=SupportsStr)\n",
    "\"\"\"Type representing the return value of the function that can be converted to string.\"\"\"\n",
    "\n",
    "\n",
    "@dataclass\n",
    "class ExceptionInfo:\n",
    "    \"\"\"Information about an exception.\"\"\"\n",
    "    name: str\n",
    "    message: str\n",
    "\n",
    "\n",
    "def validate_and_get_cast_type(__fn: Callable[..., Any]) -> type:\n",
    "    \"\"\"Validate if the function is compatible with ReAct agent and get the cast type.\"\"\"\n",
    "\n",
    "    if not inspect.isfunction(__fn):\n",
    "        raise TypeError(f\"Expected a function, got {type(__fn)}\")\n",
    "    \n",
    "    fn_params = list(inspect.signature(__fn).parameters.values())\n",
    "    if len(fn_params) != 1:\n",
    "        raise ValueError(f\"Expected a function with a single parameter, got {len(fn_params)}\")\n",
    "    \n",
    "    cast_type = fn_params[0].annotation\n",
    "    if cast_type in (Any, inspect._empty):\n",
    "        return str  # Default to string if any or no type is specified\n",
    "    \n",
    "    if cast_type not in T_Param.__constraints__:\n",
    "        constraints = tuple(map(lambda x: x.__name__, T_Param.__constraints__))\n",
    "        raise ValueError(\n",
    "            f\"Expected a function with one of the following parameter types: {constraints}\"\n",
    "        )\n",
    "    return cast_type\n",
    "        \n",
    "\n",
    "class ReActTool(Generic[T_Param, T_Retval]):\n",
    "    \"\"\"\n",
    "    Class for creating a ReAct agent compatible tool from a function.\n",
    "    \n",
    "    A ReAct agent tool is supposed to be a function that takes a single parameter.\n",
    "    \"\"\"\n",
    "    def __init__(self, __fn: Callable[[T_Param], T_Retval]):\n",
    "\n",
    "        self._cast_type = validate_and_get_cast_type(__fn)\n",
    "        \n",
    "        docstring = parse(__fn.__doc__ or \"\")\n",
    "        if docstring.description is None:\n",
    "            raise ValueError(\"Function must include a description\")\n",
    "        \n",
    "        self.__fn = __fn\n",
    "        self.name = self.__fn.__name__\n",
    "        self.description = docstring.description\n",
    "\n",
    "    def __call__(self, __arg: T_Param) -> T_Retval:\n",
    "        \"\"\"Call the function associated with the adapter.\"\"\"\n",
    "        \n",
    "        return self.__fn(__arg)\n",
    "    \n",
    "    def call_with_action_input(self, input: str) -> T_Retval | ExceptionInfo:\n",
    "        \"\"\"\n",
    "        Call the associated function with raw action input.\n",
    "        Returns `ExceptionInfo` if any exception is raised.\n",
    "        \"\"\"\n",
    "\n",
    "        try:\n",
    "            return self.__fn(self._cast_type(input))\n",
    "        except Exception as e:\n",
    "            return ExceptionInfo(name=e.__class__.__name__, message=str(e))\n",
    "\n",
    "\n",
    "def tool(__fn: Callable[[T_Param], T_Retval]) -> ReActTool[T_Param, T_Retval]:\n",
    "    \"\"\"Decorate a function to create a ReAct agent compatible tool.\"\"\"\n",
    "    return ReActTool(__fn)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implementing a ReAct agent\n",
    "\n",
    "Before we start with the implementation, let's get a basic understanding of the `ReAct` agent.\n",
    "\n",
    "`ReAct` stands for `Reason` and `Action`. It is a type of agent that processes the user's query in an iterative manner. The agent has access to a set of tools which can be used to solve the query or its sub-problems (when the agent breaks down complex queries during reasoning).\n",
    "\n",
    "The agent processes queries through the following loop:\n",
    "\n",
    "- The agent receives a query and thinks about how to solve it\n",
    "- If a tool is needed, the agent specifies:\n",
    "  - Action: The name of the tool to use\n",
    "  - Action Input: The input to provide to that tool\n",
    "- The specified tool is called and its result becomes an Observation\n",
    "- The agent then either:\n",
    "  - Provides a Final Answer if it has solved the query\n",
    "  - Or continues thinking and using more tools until it reaches a solution\n",
    "- The loop stops when either a Final Answer is provided or no valid action can be taken"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Setup Logger\n",
    "import sys\n",
    "\n",
    "from loguru import logger\n",
    "\n",
    "logger.remove()\n",
    "\n",
    "logger.add(sys.stdout, format=\"<level>{message}</level>\")\n",
    "logger = logger.opt(colors=True)\n",
    "\n",
    "def highlight_role(role: str) -> str:\n",
    "    return f\"<bg #0f0707><fg #ffffff>[{role.upper()}]</fg #ffffff></bg #0f0707>\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "import re\n",
    "from string import Template\n",
    "\n",
    "from litellm import completion as get_completion\n",
    "\n",
    "\n",
    "TOOL_TEMPLATE = Template(\"${name}: ${description}\")\n",
    "\"\"\"Template for tool description.\"\"\"\n",
    "\n",
    "\n",
    "INSTRUCTION_TEMPLATE = Template(\"\"\"${description}\n",
    "You can only utilize the following tools to answer the user's query:\n",
    "${tools_representation}\n",
    "\n",
    "Use the following format:\n",
    "\n",
    "Query: the input query you must answer\n",
    "Thought: you should always think about what to do\n",
    "Action: the action to take, should be one of ${tool_names}\n",
    "Action Input: the input to the action\n",
    "Observation: the result of the action\n",
    "... (the Thought/Action/Observation can repeat any number of times)\n",
    "Thought: What you should return as a final answer based on the last observation.\n",
    "Final Answer: The final answer to the original input query.\\\n",
    "\"\"\")\n",
    "\"\"\"Instruction template for ReAct agent.\"\"\"\n",
    "\n",
    "\n",
    "class NoResult:\n",
    "    \"\"\"A sentinel value to indicate action or tool is not available.\"\"\"\n",
    "\n",
    "\n",
    "class ReActAgent:\n",
    "    \"\"\"\n",
    "    React agent implementation that uses tools to solve given tasks.\n",
    "    The agent can be initialized with a list of functions or ReActTool instances.\n",
    "\n",
    "    This agent uses litellm to connect with Language Models.\n",
    "    Make sure to set the api key of the provider you are using.\n",
    "\n",
    "    Examples:\n",
    "        ```python\n",
    "        # Define your functions or tools\n",
    "        from duckduckgo_search import DDGS, exceptions\n",
    "        from sympy import sympify\n",
    "\n",
    "        def search_text(text: str):\n",
    "            \\\"\\\"\\\"Search for text in the web.\\\"\\\"\\\"\n",
    "            return DDGS().text(keywords=text)\n",
    "        \n",
    "        @tool   \n",
    "        def calulate_math_expression_with_sympy(expression):\n",
    "            \\\"\\\"\\\"Calculate a mathematical expression using sympy.\\\"\\\"\\\"\n",
    "            return sympify(expression)\n",
    "\n",
    "        # Initialize the agent\n",
    "        agent = ReactAgent(search_text, calulate_math_expression_with_sympy)\n",
    "\n",
    "        # Call the agent\n",
    "        agent(query=\"What is the meaning of life?\", model=\"ollama/phi4\")\n",
    "\n",
    "        # See the final instruction\n",
    "        agent.instruction\n",
    "\n",
    "        # Take a peek at the history\n",
    "        agent.show_history()\n",
    "        ```\n",
    "    \"\"\"\n",
    "    def __init__(\n",
    "        self, \n",
    "        *functions_or_tools: Callable[[T_Param], T_Retval] | ReActTool[T_Param, T_Retval],\n",
    "        default_model: str,\n",
    "        description: str | None = None,\n",
    "    ) -> None:\n",
    "        \"\"\"\n",
    "        Args:\n",
    "            functions_or_tools: A list of functions or ReActTool instances.\n",
    "            default_model: The default model to use for the agent. Explore models: https://docs.litellm.ai/docs/providers\n",
    "            description: A description that tells what an agent is and is supposed to do.\n",
    "        \"\"\"\n",
    "\n",
    "        if not functions_or_tools:\n",
    "            raise ValueError(\"Atleast one function or tool is required.\")\n",
    "\n",
    "        tools_representation = \"\"\n",
    "        self._tool_registry: dict[str, ReActTool] = {}\n",
    "        for entry in functions_or_tools:  # Update the registry with ReActTool instances\n",
    "            tool = entry if isinstance(entry, ReActTool) else ReActTool(entry)\n",
    "            self._tool_registry[tool.name] = tool\n",
    "            tool_repr = TOOL_TEMPLATE.substitute(name=tool.name, description=tool.description)\n",
    "            tools_representation += tool_repr + \"\\n\"\n",
    "        \n",
    "        self.default_model = default_model\n",
    "        description = description.strip() + \"\\n\" if description else \"\"\n",
    "        self.instruction = INSTRUCTION_TEMPLATE.substitute(\n",
    "            description=description,\n",
    "            tools_representation=tools_representation.strip(),\n",
    "            tool_names=list(self._tool_registry)\n",
    "        ).rstrip()\n",
    "        \"\"\"Instruction for the agent.\"\"\"\n",
    "\n",
    "        # Intialize message history with system instruction\n",
    "        self.__message_history = [{\"role\": \"system\", \"content\": self.instruction}]\n",
    "\n",
    "    def _parse_section(self, text: str, section: str) -> str | None:\n",
    "        \"\"\"Parse the section of the message.\"\"\"\n",
    "        _match = re.search(rf\"{section}: (.*)\", text)\n",
    "        if _match:\n",
    "            return _match.group(1).strip()\n",
    "\n",
    "    def parse_action_and_get_result(self, ai_message: str) -> str | NoResult:\n",
    "        \"\"\"Parse action and input from AI message and call the corresponding tool.\"\"\"\n",
    "\n",
    "        action = self._parse_section(ai_message, \"Action\")\n",
    "        action_input = self._parse_section(ai_message, \"Action Input\")\n",
    "        if action and action_input:\n",
    "            logger.info(\n",
    "                f\"⚙️ Action: {action!r} || ⌨️ Input: {action_input!r}\"\n",
    "            )\n",
    "            if tool := self._tool_registry.get(action):\n",
    "                return str(tool.call_with_action_input(action_input))\n",
    "        \n",
    "        return NoResult    \n",
    "\n",
    "    def process_ai_message(self, content: str) -> bool:\n",
    "        \"\"\"Process an AI message and tell if the agent should continue solving the query.\"\"\"\n",
    "\n",
    "        logger.info(highlight_role(\"self\"))\n",
    "        if thought := self._parse_section(content, \"Thought\"):\n",
    "            logger.info(f\"💭 Thought: {thought}\")\n",
    "\n",
    "        # If final answer is there, we tell the agent to stop.\n",
    "        if final := self._parse_section(content, \"Final Answer\"):\n",
    "            logger.info(f\"📌 Final Answer: {final}\")\n",
    "            self.__message_history.append({\"role\": \"assistant\", \"content\": final})\n",
    "            return False\n",
    "        \n",
    "        # Parse ai message and get the action result\n",
    "        action_result = self.parse_action_and_get_result(content)\n",
    "        if action_result is NoResult:\n",
    "            return False\n",
    "        \n",
    "        logger.info(f\"🔎 Observation: {action_result}\")\n",
    "        \n",
    "        # Add the ai message and observation to the message history\n",
    "        self.__message_history.extend([\n",
    "            {\"role\": \"assistant\", \"content\": content.strip()},\n",
    "            {\"role\": \"user\", \"content\": f\"Observation: {action_result}\"}\n",
    "        ])\n",
    "        return True\n",
    "\n",
    "    def __call__(self, *, query: str, model: str | None = None) -> str:\n",
    "        \"\"\"\n",
    "        Call the agent.\n",
    "\n",
    "        Args:\n",
    "            query: The query to complete\n",
    "            model: The model to use in the place of the default model.\n",
    "        \"\"\"\n",
    "        query = query.strip()\n",
    "        logger.info(f\"🖋️ Query: {query}\\n\")\n",
    "\n",
    "        # Add user query to message history\n",
    "        self.__message_history.append({\"role\": \"user\", \"content\": f\"Query: {query}\"})\n",
    "\n",
    "        while True:\n",
    "            # Get completion response from the model\n",
    "            ai_message = get_completion(\n",
    "                model or self.default_model, self.__message_history, stop=[\"Observation:\"]\n",
    "            ).choices[0].message.content\n",
    "            \n",
    "            # Process the AI message\n",
    "            if not self.process_ai_message(ai_message):\n",
    "                break\n",
    "            \n",
    "            logger.info(\"\")\n",
    "        \n",
    "    @property\n",
    "    def history(self) -> list[dict[str, str]]:\n",
    "        \"\"\"Messasge history\"\"\"\n",
    "        return self.__message_history\n",
    "\n",
    "    def show_history(self) -> None:\n",
    "        \"\"\"Display the message history.\"\"\"\n",
    "        for message in self.__message_history:\n",
    "            logger.info(f\"{highlight_role(message['role'])} \\n{message['content']}\\n\")\n",
    "\n",
    "    def clear_history(self) -> None:\n",
    "        \"\"\"Clear the message history.\"\"\"\n",
    "        self.__message_history = [{\"role\": \"system\", \"content\": self.instruction}]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Creating a ReAct agent\n",
    "\n",
    "I am using Groq because it is super fast and has a good free usage limit. Get your own key from [here](https://console.groq.com/keys). If you want to use other models, you can select one from the [LiteLLM providers](https://docs.litellm.ai/docs/providers)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "from getpass import getpass\n",
    "\n",
    "def set_env(var: str):\n",
    "    if not os.environ.get(var):\n",
    "        os.environ[var] = getpass(f\"{var}: \")\n",
    "\n",
    "set_env(\"GROQ_API_KEY\")  # Replace with your LLM compatible environment key."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "You can only utilize the following tools to answer the user's query:\n",
      "search_text: Search for text in the web.\n",
      "calulate_math_expression_with_sympy: Calculate a mathematical expression using sympy.\n",
      "\n",
      "Use the following format:\n",
      "\n",
      "Query: the input query you must answer\n",
      "Thought: you should always think about what to do\n",
      "Action: the action to take, should be one of ['search_text', 'calulate_math_expression_with_sympy']\n",
      "Action Input: the input to the action\n",
      "Observation: the result of the action\n",
      "... (the Thought/Action/Observation can repeat any number of times)\n",
      "Thought: What you should return as a final answer based on the last observation.\n",
      "Final Answer: The final answer to the original input query.\n"
     ]
    }
   ],
   "source": [
    "from duckduckgo_search import DDGS\n",
    "from sympy import sympify\n",
    "\n",
    "def search_text(text: str):\n",
    "    \"\"\"Search for text in the web.\"\"\"\n",
    "    \n",
    "    return DDGS().text(keywords=text)\n",
    "\n",
    "## Decorating is not really required, normal functions will work fine. \n",
    "## I am just trying to keep things flexible.\n",
    "@tool\n",
    "def calulate_math_expression_with_sympy(expression):\n",
    "    \"\"\"Calculate a mathematical expression using sympy.\"\"\"\n",
    "\n",
    "    return sympify(expression)\n",
    "\n",
    "agent = ReActAgent(\n",
    "    search_text, calulate_math_expression_with_sympy, default_model=\"groq/llama-3.3-70b-versatile\"\n",
    ")\n",
    "print(agent.instruction)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m🖋️ Query: Find some information about smolagents library.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: To find information about the smolagents library, I should first search for it on the web to see if there are any relevant results.\u001b[0m\n",
      "\u001b[1m⚙️ Action: 'search_text' || ⌨️ Input: 'smolagents library'\u001b[0m\n",
      "\u001b[1m🔎 Observation: [{'title': 'smolagents: a barebones library for agents. Agents write ... - GitHub', 'href': 'https://github.com/huggingface/smolagents', 'body': 'smolagents is a library that enables you to run powerful agents in a few lines of code. It offers: Simplicity: the logic for agents fits in ~thousand lines of code (see agents.py).We kept abstractions to their minimal shape above raw code! 🧑\\u200d💻 First-class support for Code Agents, i.e. agents that write their actions in code (as opposed to \"agents being used to write code\").'}, {'title': 'Smolagents : Huggingface AI Agent Framework', 'href': 'https://smolagents.org/', 'body': 'smolagents is a minimalist AI agent library developed by the Hugging Face team. It allows developers to create and run powerful AI agents with minimal code. By focusing on simplicity and efficiency, smolagents enables large language models (LLMs) to interact seamlessly with real-world tasks.'}, {'title': 'smolagents - Hugging Face', 'href': 'https://huggingface.co/docs/smolagents/index', 'body': 'smolagents. This library is the simplest framework out there to build powerful agents! By the way, wtf are \"agents\"? We provide our definition in this page, where you\\'ll also find tips for when to use them or not (spoilers: you\\'ll often be better off without agents).. This library offers:'}, {'title': 'Introducing smolagents: A Lightweight Library for Building ... - Medium', 'href': 'https://medium.com/thedeephub/introducing-smolagents-a-lightweight-library-for-building-powerful-agents-a8791a60b5b1', 'body': 'smolagents is an open-source project backed by Hugging Face, fostering a community of contributors and users. The library integrates seamlessly with the Hugging Face Hub, allowing you to share and ...'}, {'title': 'Smolagents - Smolagents', 'href': 'https://smolagents.org/docs/smolagent-docs/', 'body': 'This library is the simplest framework out there to build powerful agents! By the way, wtf are \"agents\"? We provide our definition in this page, where you\\'ll also find tips for when to use them or not (spoilers: you\\'ll often be better off without agents). This library offers: Simplicity: the logic for agents fits in ~thousand...'}, {'title': 'Introduction to Smolagents: A Hugging Face Agentic Framework', 'href': 'https://medium.com/@mauryaanoop3/introduction-to-smolagents-a-hugging-face-agentic-framework-190169b424f4', 'body': 'What Are Smolagents? Smolagents is a minimalist yet powerful library that enables the construction of AI agents with minimal code. By providing a simplified interface, it allows developers to ...'}, {'title': 'SmolAgents by Hugging Face: Build AI Agents in Under 30 Lines', 'href': 'https://www.analyticsvidhya.com/blog/2025/01/smolagents/', 'body': \"The SmolAgents library systematically refined the code across iterations to handle errors, improve robustness, and achieve the desired result of fetching Apple Inc.'s stock price (246.21). Conclusion. SmolAgents by Hugging Face represents a groundbreaking advancement in simplifying the creation and execution of AI-powered agents. With its ...\"}, {'title': 'Introduction to Agents - Hugging Face', 'href': 'https://huggingface.co/docs/smolagents/conceptual_guides/intro_agents', 'body': 'smolagents documentation Introduction to Agents. smolagents Search documentation. Get started. 🤗 Agents Guided tour. Tutorials. Building good agents 🛠️ Tools - in-depth guide 🛡️ Secure your code execution with E2B. Conceptual guides. 🤖 An introduction to agentic systems 🤔 How do Multi-step agents work? ...'}, {'title': 'Agents - Guided tour - Hugging Face', 'href': 'https://huggingface.co/docs/smolagents/guided_tour', 'body': 'The system prompt includes: An introduction that explains how the agent should behave and what tools are.; A description of all the tools that is defined by a {{tool_descriptions}} token that is dynamically replaced at runtime with the tools defined/chosen by the user.. The tool description comes from the tool attributes, name, description, inputs and output_type, and a simple jinja2 template ...'}, {'title': 'Install and Run Powerful smolagents library and AI Agents by Using ...', 'href': 'https://aleksandarhaber.com/install-and-run-powerful-smolagents-library-and-ai-agents-by-using-ollama-and-llama/', 'body': 'Instead, we will explain how to install and use smolagents library \"locally\" by using Ollama and Llama 3.2B. Once you learn how to install smolagents and run test examples, you can proceed further and learn how to develop your own AI agents. You can also use the ideas presented in this tutorial to run any other Large Language Model (LLM ...'}]\u001b[0m\n",
      "\u001b[1m\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: The search results provide a wealth of information about the smolagents library, including its features, how it works, and how to use it. I can see that it's a minimalist AI agent library developed by the Hugging Face team, and it allows developers to create and run powerful AI agents with minimal code.\u001b[0m\n",
      "\u001b[1m📌 Final Answer: The smolagents library is a minimalist AI agent library developed by the Hugging Face team, allowing developers to create and run powerful AI agents with minimal code. It offers simplicity, first-class support for code agents, and seamless integration with large language models, and is open-source and backed by Hugging Face.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "agent(query=\"Find some information about smolagents library.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m🖋️ Query: Calculate square root of 144.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: To calculate the square root of 144, I can use a mathematical expression and solve it using a calculator or a programming language. Since I have access to the sympy library, I can use it to calculate the square root.\u001b[0m\n",
      "\u001b[1m⚙️ Action: 'calulate_math_expression_with_sympy' || ⌨️ Input: 'sqrt(144)'\u001b[0m\n",
      "\u001b[1m🔎 Observation: 12\u001b[0m\n",
      "\u001b[1m\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: The sympy library has calculated the square root of 144, which is 12. This is the correct result, as 12 * 12 = 144.\u001b[0m\n",
      "\u001b[1m📌 Final Answer: The square root of 144 is 12.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "agent(query=\"Calculate square root of 144.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SYSTEM]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "\n",
      "You can only utilize the following tools to answer the user's query:\n",
      "search_text: Search for text in the web.\n",
      "calulate_math_expression_with_sympy: Calculate a mathematical expression using sympy.\n",
      "\n",
      "Use the following format:\n",
      "\n",
      "Query: the input query you must answer\n",
      "Thought: you should always think about what to do\n",
      "Action: the action to take, should be one of ['search_text', 'calulate_math_expression_with_sympy']\n",
      "Action Input: the input to the action\n",
      "Observation: the result of the action\n",
      "... (the Thought/Action/Observation can repeat any number of times)\n",
      "Thought: What you should return as a final answer based on the last observation.\n",
      "Final Answer: The final answer to the original input query.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[USER]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Query: Find some information about smolagents library.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[ASSISTANT]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Thought: To find information about the smolagents library, I should first search for it on the web to see if there are any relevant results.\n",
      "\n",
      "Action: search_text\n",
      "Action Input: smolagents library\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[USER]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Observation: [{'title': 'smolagents: a barebones library for agents. Agents write ... - GitHub', 'href': 'https://github.com/huggingface/smolagents', 'body': 'smolagents is a library that enables you to run powerful agents in a few lines of code. It offers: Simplicity: the logic for agents fits in ~thousand lines of code (see agents.py).We kept abstractions to their minimal shape above raw code! 🧑\\u200d💻 First-class support for Code Agents, i.e. agents that write their actions in code (as opposed to \"agents being used to write code\").'}, {'title': 'Smolagents : Huggingface AI Agent Framework', 'href': 'https://smolagents.org/', 'body': 'smolagents is a minimalist AI agent library developed by the Hugging Face team. It allows developers to create and run powerful AI agents with minimal code. By focusing on simplicity and efficiency, smolagents enables large language models (LLMs) to interact seamlessly with real-world tasks.'}, {'title': 'smolagents - Hugging Face', 'href': 'https://huggingface.co/docs/smolagents/index', 'body': 'smolagents. This library is the simplest framework out there to build powerful agents! By the way, wtf are \"agents\"? We provide our definition in this page, where you\\'ll also find tips for when to use them or not (spoilers: you\\'ll often be better off without agents).. This library offers:'}, {'title': 'Introducing smolagents: A Lightweight Library for Building ... - Medium', 'href': 'https://medium.com/thedeephub/introducing-smolagents-a-lightweight-library-for-building-powerful-agents-a8791a60b5b1', 'body': 'smolagents is an open-source project backed by Hugging Face, fostering a community of contributors and users. The library integrates seamlessly with the Hugging Face Hub, allowing you to share and ...'}, {'title': 'Smolagents - Smolagents', 'href': 'https://smolagents.org/docs/smolagent-docs/', 'body': 'This library is the simplest framework out there to build powerful agents! By the way, wtf are \"agents\"? We provide our definition in this page, where you\\'ll also find tips for when to use them or not (spoilers: you\\'ll often be better off without agents). This library offers: Simplicity: the logic for agents fits in ~thousand...'}, {'title': 'Introduction to Smolagents: A Hugging Face Agentic Framework', 'href': 'https://medium.com/@mauryaanoop3/introduction-to-smolagents-a-hugging-face-agentic-framework-190169b424f4', 'body': 'What Are Smolagents? Smolagents is a minimalist yet powerful library that enables the construction of AI agents with minimal code. By providing a simplified interface, it allows developers to ...'}, {'title': 'SmolAgents by Hugging Face: Build AI Agents in Under 30 Lines', 'href': 'https://www.analyticsvidhya.com/blog/2025/01/smolagents/', 'body': \"The SmolAgents library systematically refined the code across iterations to handle errors, improve robustness, and achieve the desired result of fetching Apple Inc.'s stock price (246.21). Conclusion. SmolAgents by Hugging Face represents a groundbreaking advancement in simplifying the creation and execution of AI-powered agents. With its ...\"}, {'title': 'Introduction to Agents - Hugging Face', 'href': 'https://huggingface.co/docs/smolagents/conceptual_guides/intro_agents', 'body': 'smolagents documentation Introduction to Agents. smolagents Search documentation. Get started. 🤗 Agents Guided tour. Tutorials. Building good agents 🛠️ Tools - in-depth guide 🛡️ Secure your code execution with E2B. Conceptual guides. 🤖 An introduction to agentic systems 🤔 How do Multi-step agents work? ...'}, {'title': 'Agents - Guided tour - Hugging Face', 'href': 'https://huggingface.co/docs/smolagents/guided_tour', 'body': 'The system prompt includes: An introduction that explains how the agent should behave and what tools are.; A description of all the tools that is defined by a {{tool_descriptions}} token that is dynamically replaced at runtime with the tools defined/chosen by the user.. The tool description comes from the tool attributes, name, description, inputs and output_type, and a simple jinja2 template ...'}, {'title': 'Install and Run Powerful smolagents library and AI Agents by Using ...', 'href': 'https://aleksandarhaber.com/install-and-run-powerful-smolagents-library-and-ai-agents-by-using-ollama-and-llama/', 'body': 'Instead, we will explain how to install and use smolagents library \"locally\" by using Ollama and Llama 3.2B. Once you learn how to install smolagents and run test examples, you can proceed further and learn how to develop your own AI agents. You can also use the ideas presented in this tutorial to run any other Large Language Model (LLM ...'}]\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[ASSISTANT]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "The smolagents library is a minimalist AI agent library developed by the Hugging Face team, allowing developers to create and run powerful AI agents with minimal code. It offers simplicity, first-class support for code agents, and seamless integration with large language models, and is open-source and backed by Hugging Face.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[USER]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Query: Calculate square root of 144.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[ASSISTANT]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Thought: To calculate the square root of 144, I can use a mathematical expression and solve it using a calculator or a programming language. Since I have access to the sympy library, I can use it to calculate the square root.\n",
      "\n",
      "Action: calulate_math_expression_with_sympy\n",
      "Action Input: sqrt(144)\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[USER]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Observation: 12\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[ASSISTANT]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "The square root of 144 is 12.\n",
      "\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "agent.show_history()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Creating a role-playing ReAct agent\n",
    "\n",
    "We can pass custom description when creating an agent. This will tell the agent its purpose of using the tools."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "You are a research agent who makes use of tools to find information and answer user queries.\n",
      "\n",
      "You can only utilize the following tools to answer the user's query:\n",
      "search_text: Search for text in the web.\n",
      "get_weather: Get weather data for a given city.\n",
      "\n",
      "Use the following format:\n",
      "\n",
      "Query: the input query you must answer\n",
      "Thought: you should always think about what to do\n",
      "Action: the action to take, should be one of ['search_text', 'get_weather']\n",
      "Action Input: the input to the action\n",
      "Observation: the result of the action\n",
      "... (the Thought/Action/Observation can repeat any number of times)\n",
      "Thought: What you should return as a final answer based on the last observation.\n",
      "Final Answer: The final answer to the original input query.\n"
     ]
    }
   ],
   "source": [
    "import requests\n",
    "from dataclasses import dataclass\n",
    "from typing import Any\n",
    "\n",
    "def fetch_json(url: str) -> dict[str, Any]:\n",
    "    \"\"\"Fetch JSON data from a given URL.\"\"\"\n",
    "    \n",
    "    response = requests.get(url)\n",
    "    response.raise_for_status()\n",
    "    return response.json()\n",
    "\n",
    "@dataclass\n",
    "class WeatherData:\n",
    "    temperature: float\n",
    "    wind_speed: float\n",
    "\n",
    "    def __str__(self) -> str:\n",
    "        return f\"Current temperature: {self.temperature}°C, Current wind speed: {self.wind_speed} km/h\"\n",
    "\n",
    "def get_weather(city_name: str) -> WeatherData:\n",
    "    \"\"\"Get weather data for a given city.\"\"\"\n",
    "\n",
    "    json_response = fetch_json(\n",
    "        f\"https://geocoding-api.open-meteo.com/v1/search?name={city_name}\"\n",
    "    )\n",
    "\n",
    "    if results := json_response.get(\"results\", []):\n",
    "        latitude, longitude = results[0][\"latitude\"], results[0][\"longitude\"]\n",
    "    else:\n",
    "        raise LookupError(f\"City {city_name!r} not found.\")\n",
    "    \n",
    "    current_weather = fetch_json(\n",
    "        f\"https://api.open-meteo.com/v1/forecast?\"\n",
    "        f\"latitude={latitude}&longitude={longitude}&current_weather=true\"\n",
    "    ).get(\"current_weather\", {})\n",
    "\n",
    "    try:\n",
    "        return WeatherData(\n",
    "            temperature=current_weather[\"temperature\"],\n",
    "            wind_speed=current_weather[\"windspeed\"],\n",
    "        )\n",
    "    except KeyError:\n",
    "        raise LookupError(\"Weather data not available for city {city_name!r}.\")\n",
    "\n",
    "research_agent = ReActAgent(\n",
    "    search_text, get_weather, \n",
    "    default_model=\"groq/llama-3.3-70b-versatile\",\n",
    "    description=\"\"\"\\\n",
    "You are a research agent who makes use of tools to find information and answer user queries.\n",
    "\"\"\"\n",
    ")\n",
    "print(research_agent.instruction)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m🖋️ Query: Based on the current weather of Kolkata, find the best place to visit.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: To find the best place to visit in Kolkata based on the current weather, I need to first get the current weather data for Kolkata.\u001b[0m\n",
      "\u001b[1m⚙️ Action: 'get_weather' || ⌨️ Input: 'Kolkata'\u001b[0m\n",
      "\u001b[1m🔎 Observation: Current temperature: 25.9°C, Current wind speed: 8.9 km/h\u001b[0m\n",
      "\u001b[1m\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: With a current temperature of 25.9°C and a wind speed of 8.9 km/h, the weather in Kolkata is relatively pleasant. Now, I need to find out the best places to visit in Kolkata that are suitable for this kind of weather.\u001b[0m\n",
      "\u001b[1m⚙️ Action: 'search_text' || ⌨️ Input: 'best places to visit in Kolkata in pleasant weather'\u001b[0m\n",
      "\u001b[1m🔎 Observation: [{'title': '13 Best Places to Visit in Kolkata (Calcutta) in ... - BharatTravelGuru', 'href': 'https://bharattravelguru.com/blog/places-to-visit-in-kolkata/', 'body': 'The best time to visit Bakkhali is from October to March when the weather is pleasant. 2. Digha: ... One thought on \"13 Best Places to Visit in Kolkata (Calcutta) in 2023: How to Reach & Best Time to Visit\" Aadhya says: 25 November 2023 at 09:36. Nice article. Very detailed information provided….'}, {'title': 'The Best Time to Visit Kolkata - TripSavvy', 'href': 'https://www.tripsavvy.com/best-time-to-visit-kolkata-5113998', 'body': \"Weather in Kolkata . The great thing about Kolkata's tropical climate is that the city never gets really cold. In winter, temperatures range from around 54 to 79 degrees F (12 to 26 degrees C), with nippy nights but sunny days. However, pollution becomes trapped in the atmosphere, resulting in poor air quality in the city.\"}, {'title': 'Best Time to Visit in Kolkata | Temperature, Weather & Seasons - MakeMyTrip', 'href': 'https://www.makemytrip.com/tripideas/kolkata/best-time-to-visit', 'body': 'Ideally the best time to visit any place is considered when the weather conditions are delightful. For Kolkata, you can choose to go in October, November, December, January, February and March which is the peak season and April and May which is the moderate season. Both will serve you well to unwind from your tedious routine.'}, {'title': 'Visit the best places in Kolkata in Winter - niccoparks', 'href': 'https://www.niccoparks.com/best-places-to-visit-in-kolkata-during-winter/', 'body': \"Kolkata, the City of Joy, becomes even more delightful during winter, with pleasant weather and a festive atmosphere. The city offers a perfect blend of history, culture, and natural beauty, making it an ideal destination for winter travelers. Here's a curated list of must-visit places in Kolkata during the winter in kolkata.\"}, {'title': '10 Best Places To Visit In Kolkata, India [2025] Photos', 'href': 'https://fierytrippers.com/best-places-to-visit-in-kolkata/', 'body': \"Best Places To Visit In Kolkata: 1. Victoria Memorial Structure, 2. Indian Museum, 3. Birla Planetarium, 4. St. Paul's Cathedral. ... (October to February) for pleasant weather. Embrace local transport like trams and hand-pulled rickshaws for a unique experience. Best Seasons:\"}, {'title': '52 Best Places to visit in Kolkata | Top Tourist Attractions | 2025', 'href': 'https://www.holidify.com/places/kolkata/sightseeing-and-things-to-do.html', 'body': 'See most popular tourist places to visit in Kolkata, top things to do, shopping and nightlife in Kolkata, find entry timings, fees about various attractions in Kolkata, West Bengal ... The water park occupies over 75000 square feet of area and provides for a pleasant respite from the scorching summer heat. They provide lockers for the visitors ...'}, {'title': 'Kolkata Weather > Best Time To Visit Calcutta, Temperature - Holidify', 'href': 'https://www.holidify.com/places/kolkata/best-time-to-visit.html', 'body': 'The autumn and winter months between October and February are the best time to visit Kolkata, as the weather is perfect and festivities are aplenty. This is the season which witnesses the peak arrival of travellers, especially during Durga Puja. ... You can take a walk by the Hooghly River in the evenings and have pleasant outings as well. Take ...'}, {'title': 'Amazing & Beautiful 10 Best Places To Visit In Kolkata', 'href': 'https://trueyatra.com/explore-india/best-places-to-visit-in-kolkata/', 'body': '10. Science City, Kolkata . This is one of the best places to visit in Kolkata and also a popular tourist destination which is visited by a large number of tourists every day. The major highlights of this place include Space Theatre, 3D shows, Natural Science Park, Evolution Theme Park, Dinosaurs Complex, and Time Machine.'}, {'title': 'Places to Visit in Kolkata - MakeMyTrip', 'href': 'https://www.makemytrip.com/tripideas/places-to-visit-in-kolkata', 'body': \"Whattoexpect:Cool weather. Thingsyou'lllove:Regional festivals, street food and sightseeing. FESTIVALS AND EVENTS. 1 / 2. Ramzan As per Islamic Calendar. ... This is the perfect time to explore the best places to visit in Kolkata since the temperature is just right and the weather is really lovely. Places to visit in Kolkata with family & friends.\"}, {'title': '33 Best Places to Visit in Kolkata: Ultimate Travel Guide - WanderOn', 'href': 'https://wanderon.in/blogs/places-to-visit-in-kolkata', 'body': \"Let's embark on a journey through the top 33 places to visit in Kolkata, each promising a unique experience and unforgettable memories. you can find the best tourist attractions, food destinations, family-friendly spots, and shopping hubs in Kolkata with this travel guide. Suggested Read: Rain-Soaked Wonders: Monsoon In West Bengal\"}]\u001b[0m\n",
      "\u001b[1m\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: Based on the search results, the best places to visit in Kolkata in pleasant weather include Victoria Memorial Structure, Indian Museum, Birla Planetarium, St. Paul's Cathedral, Science City, and many more. These places offer a mix of history, culture, and entertainment, making them ideal for visitors.\u001b[0m\n",
      "\u001b[1m📌 Final Answer: The best places to visit in Kolkata based on the current weather are Victoria Memorial Structure, Indian Museum, Birla Planetarium, St. Paul's Cathedral, and Science City. These destinations offer a unique blend of history, culture, and entertainment, making them perfect for exploring the city in pleasant weather.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "research_agent(\n",
    "    query=\"Based on the current weather of Kolkata, find the best place to visit.\"\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### How does this agent handles exceptions?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "def search_web(query: str) -> str:\n",
    "    \"\"\"Search the web for the given query.\"\"\"\n",
    "    raise ConnectionError(\"No Internet connection.\")\n",
    "\n",
    "agent = ReActAgent(\n",
    "    search_web, default_model=\"groq/llama-3.3-70b-versatile\"\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m🖋️ Query: Find me some good tutorials on how to get started with Python.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: To find good tutorials on getting started with Python, I should search the web for relevant resources.\u001b[0m\n",
      "\u001b[1m⚙️ Action: 'search_web' || ⌨️ Input: 'Python tutorials for beginners'\u001b[0m\n",
      "\u001b[1m🔎 Observation: ExceptionInfo(name='ConnectionError', message='No Internet connection.')\u001b[0m\n",
      "\u001b[1m\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SELF]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m\u001b[0m\n",
      "\u001b[1m💭 Thought: It seems I'm experiencing a connection error, which is preventing me from searching the web for Python tutorials. I'll have to rely on my prior knowledge to provide some general information on where to find Python tutorials.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "agent(query=\"Find me some good tutorials on how to get started with Python.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[SYSTEM]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "\n",
      "You can only utilize the following tools to answer the user's query:\n",
      "search_web: Search the web for the given query.\n",
      "\n",
      "Use the following format:\n",
      "\n",
      "Query: the input query you must answer\n",
      "Thought: you should always think about what to do\n",
      "Action: the action to take, should be one of ['search_web']\n",
      "Action Input: the input to the action\n",
      "Observation: the result of the action\n",
      "... (the Thought/Action/Observation can repeat any number of times)\n",
      "Thought: What you should return as a final answer based on the last observation.\n",
      "Final Answer: The final answer to the original input query.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[USER]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Query: Find me some good tutorials on how to get started with Python.\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[ASSISTANT]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Thought: To find good tutorials on getting started with Python, I should search the web for relevant resources.\n",
      "\n",
      "Action: search_web\n",
      "Action Input: Python tutorials for beginners\n",
      "\u001b[0m\n",
      "\u001b[1m\u001b[48;2;15;7;7m\u001b[38;2;255;255;255m[USER]\u001b[0m\u001b[1m\u001b[48;2;15;7;7m\u001b[0m\u001b[1m \n",
      "Observation: ExceptionInfo(name='ConnectionError', message='No Internet connection.')\n",
      "\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "agent.show_history()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}