notebooks/react_agent.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ReACT Agent\n",
    "\n",
    "For many, a Large Language Model (LLM) Agent is synonymous with the ReACT architecture. This simple yet powerful architecture is present in most mainstream Agentic frameworks. In this notebook, I aim to explain how ReACT works under the hood and demonstrate that HybridAGI, as a general-purpose, programmable graph-based Agentic framework, can indeed build a ReACT agent.\n",
    "\n",
    "However, we do not advocate for the ReACT architecture as the ultimate solution, especially for complex tasks, due to its lack of control and efficiency. Nevertheless, if you're new to Graph-based Prompt Programming, a ReACT agent can serve as an excellent starting point. With HybridAGI, you can iteratively enhance this initial agent and deviate from the ReACT architecture by adding more actions and decisions, all within the same framework. This flexibility is one of the key strengths of HybridAGI.\n",
    "\n",
    "In its essence, a ReACT agent is composed by two elements:\n",
    "\n",
    "- A decision for choosing the tool to use\n",
    "- A loop to iteratively adapt to the user input"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/yoan/.cache/pypoetry/virtualenvs/hybridagi-B1GoJrSC-py3.10/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "// @desc: The main program\n",
      "CREATE\n",
      "// Nodes declaration\n",
      "(start:Control {id: \"start\"}),\n",
      "(end:Control {id: \"end\"}),\n",
      "(tool_choice:Decision {\n",
      "  id: \"tool_choice\",\n",
      "  purpose: \"Choose the next tool to use\",\n",
      "  question: \"Which tool to use for the next step?\\nUse the context to help you choose.\\nTo give the final answer just select finish\"\n",
      "}),\n",
      "(ask_user:Action {\n",
      "  id: \"ask_user\",\n",
      "  purpose: \"Ask the user\",\n",
      "  tool: \"AskUser\",\n",
      "  prompt: \"Ask a question to the user\"\n",
      "}),\n",
      "(finish:Action {\n",
      "  id: \"finish\",\n",
      "  purpose: \"End the conversation and give the final answer\",\n",
      "  tool: \"Speak\",\n",
      "  prompt: \"Please give the final answer, if you don't know just say that you don't know\"\n",
      "}),\n",
      "// Structure declaration\n",
      "(start)-[:NEXT]->(tool_choice),\n",
      "(tool_choice)-[:ASK_USER]->(ask_user),\n",
      "(tool_choice)-[:FINISH]->(finish),\n",
      "(ask_user)-[:NEXT]->(tool_choice),\n",
      "(finish)-[:NEXT]->(end)\n"
     ]
    }
   ],
   "source": [
    "# Let's start implementing it!\n",
    "\n",
    "import hybridagi.core.graph_program as gp\n",
    "\n",
    "main = gp.GraphProgram(\n",
    "    name = \"main\",\n",
    "    description = \"The main program\",\n",
    ")\n",
    "\n",
    "main.add(gp.Decision(\n",
    "    id = \"tool_choice\",\n",
    "    purpose = \"Choose the next tool to use\",\n",
    "    question = \\\n",
    "\"\"\"Which tool to use for the next step?\n",
    "Use the context to help you choose.\n",
    "To give the final answer just select finish\"\"\",\n",
    "))\n",
    "\n",
    "main.add(gp.Action(\n",
    "    id = \"ask_user\",\n",
    "    purpose = \"Ask the user\",\n",
    "    tool = \"AskUser\",\n",
    "    prompt = \"Ask a question to the user\",\n",
    "))\n",
    "\n",
    "main.add(gp.Action(\n",
    "    id = \"finish\",\n",
    "    purpose = \"End the conversation and give the final answer\",\n",
    "    tool = \"Speak\",\n",
    "    prompt = \"Please give the final answer, if you don't know just say that you don't know\",\n",
    "))\n",
    "\n",
    "main.connect(\"start\", \"tool_choice\")\n",
    "main.connect(\"tool_choice\", \"ask_user\", label=\"Ask User\")\n",
    "main.connect(\"tool_choice\", \"finish\", label=\"Finish\")\n",
    "main.connect(\"ask_user\", \"tool_choice\")\n",
    "main.connect(\"finish\", \"end\")\n",
    "\n",
    "main.build()\n",
    "\n",
    "print(main)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[35m--- Step 0 ---\n",
      "Call Program: main\n",
      "Program Purpose: Please teach me how to how to bake a chocolate, vanilla or strawberry cake\u001b[0m\n",
      "\u001b[34m--- Step 1 ---\n",
      "Decision Purpose: Choose the next tool to use\n",
      "Decision Question: Which tool to use for the next step?\n",
      "Use the context to help you choose.\n",
      "To give the final answer just select finish\n",
      "Decision: ASK_USER\u001b[0m\n",
      "\u001b[36m--- Step 2 ---\n",
      "Action Purpose: Ask the user\n",
      "Action: {\n",
      "  \"question\": \"What flavor of cake would you like to bake, chocolate, vanilla or strawberry?\",\n",
      "  \"answer\": \"Chocolate cake, please.\"\n",
      "}\u001b[0m\n",
      "\u001b[34m--- Step 3 ---\n",
      "Decision Purpose: Choose the next tool to use\n",
      "Decision Question: Which tool to use for the next step?\n",
      "Use the context to help you choose.\n",
      "To give the final answer just select finish\n",
      "Decision: ASK_USER\u001b[0m\n",
      "\u001b[36m--- Step 4 ---\n",
      "Action Purpose: Ask the user\n",
      "Action: {\n",
      "  \"question\": \"What flavor of cake would you like to bake, chocolate, vanilla or strawberry?\",\n",
      "  \"answer\": \"Chocolate cake, as previously stated.\"\n",
      "}\u001b[0m\n",
      "\u001b[34m--- Step 5 ---\n",
      "Decision Purpose: Choose the next tool to use\n",
      "Decision Question: Which tool to use for the next step?\n",
      "Use the context to help you choose.\n",
      "To give the final answer just select finish\n",
      "Decision: FINISH\u001b[0m\n",
      "\u001b[36m--- Step 6 ---\n",
      "Action Purpose: End the conversation and give the final answer\n",
      "Action: {\n",
      "  \"message\": \"To bake a chocolate cake, here are the steps you need to follow:\\n\\n1. Preheat your oven to 350\\u00b0F (175\\u00b0C). Grease and flour a 9-inch round baking pan.\\n\\n2. In a large bowl, combine 1 3/4 cups all-purpose flour, 2 teaspoons baking powder, 1/2 teaspoon baking soda, 1/2 teaspoon salt, and 3/4 cup unsweetened cocoa powder. Set aside.\\n\\n3. In another large bowl, beat 1 1/2 cups sugar with 2 eggs until light and fluffy. Add 1 cup milk, 1/2 cup vegetable oil, 2 teaspoons vanilla extract, and 1 cup boiling water to the egg mixture, beating constantly.\\n\\n4. Gradually add the flour mixture to the egg mixture, beating well after each addition. Pour the batter into the prepared pan.\\n\\n5. Bake for 30-35 minutes or until a toothpick inserted in the center comes out clean. Cool in the pan for 10 minutes on a wire rack; remove from the pan and cool completely.\\n\\n6. Once your chocolate cake is baked, you can frost it with your favorite frosting such as chocolate ganache or buttercream. Enjoy your homemade chocolate cake!\"\n",
      "}\u001b[0m\n",
      "\u001b[35m--- Step 7 ---\n",
      "End Program: main\u001b[0m\n",
      "To bake a chocolate cake, here are the steps you need to follow:\n",
      "\n",
      "1. Preheat your oven to 350°F (175°C). Grease and flour a 9-inch round baking pan.\n",
      "\n",
      "2. In a large bowl, combine 1 3/4 cups all-purpose flour, 2 teaspoons baking powder, 1/2 teaspoon baking soda, 1/2 teaspoon salt, and 3/4 cup unsweetened cocoa powder. Set aside.\n",
      "\n",
      "3. In another large bowl, beat 1 1/2 cups sugar with 2 eggs until light and fluffy. Add 1 cup milk, 1/2 cup vegetable oil, 2 teaspoons vanilla extract, and 1 cup boiling water to the egg mixture, beating constantly.\n",
      "\n",
      "4. Gradually add the flour mixture to the egg mixture, beating well after each addition. Pour the batter into the prepared pan.\n",
      "\n",
      "5. Bake for 30-35 minutes or until a toothpick inserted in the center comes out clean. Cool in the pan for 10 minutes on a wire rack; remove from the pan and cool completely.\n",
      "\n",
      "6. Once your chocolate cake is baked, you can frost it with your favorite frosting such as chocolate ganache or buttercream. Enjoy your homemade chocolate cake!\n"
     ]
    }
   ],
   "source": [
    "# Now we setup our agent\n",
    "\n",
    "import dspy\n",
    "from hybridagi.core.datatypes import AgentState, Query\n",
    "from hybridagi.modules.agents import GraphInterpreterAgent\n",
    "from hybridagi.memory.integration.local import LocalProgramMemory\n",
    "from hybridagi.modules.agents.tools import (\n",
    "    SpeakTool,\n",
    "    AskUserTool,\n",
    ")\n",
    "\n",
    "program_memory = LocalProgramMemory(index_name=\"react_agent\")\n",
    "\n",
    "# Never forget to load your main program into memory!\n",
    "\n",
    "program_memory.update(main)\n",
    "\n",
    "# Setup the tools\n",
    "\n",
    "agent_state = AgentState()\n",
    "\n",
    "tools = [\n",
    "    SpeakTool(\n",
    "        agent_state = agent_state,\n",
    "    ),\n",
    "    AskUserTool(\n",
    "        agent_state = agent_state,\n",
    "    ),\n",
    "]\n",
    "\n",
    "agent = GraphInterpreterAgent(\n",
    "    program_memory = program_memory,\n",
    "    agent_state = agent_state,\n",
    "    tools = tools,\n",
    ")\n",
    "\n",
    "# Don't forget to setup you LLM\n",
    "\n",
    "lm = dspy.OllamaLocal(model='mistral', max_tokens=1024, stop=[\"\\n\\n\\n\", \"\\n---\"])\n",
    "dspy.configure(lm=lm)\n",
    "\n",
    "result = agent(Query(query=\"Please teach me how to how to bake a chocolate, vanilla or strawberry cake\"))\n",
    "\n",
    "print(result.final_answer)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So next time that you see a ReACT agent, you should have this structure in mind. And you should know why they are not the future of Agentic systems and why you should forget about any framework that emphasize their use."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "hybridagi-B1GoJrSC-py3.10",
   "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}