Looking to build intelligent agents with real-world capabilities? Use Google ADK for building agents that can reason, delegate, and respond dynamically. This Google ADK tutorial walks you through the steps to build conversational agents with Google ADK across different language models like Gemini<\/a> and GPT<\/a>. Whether you\u2019re exploring Google ADK for AI agents or curious about how to create AI agents<\/a> using Google ADK, this hands-on guide will help you kickstart your journey into agentic development with ease and clarity.<\/p>\n Agent Development Kit (ADK) is a flexible and modular framework for developing and deploying AI agents. It can be used with popular LLMs and open-source generative AI tools and is designed to integrate tightly with the Google ecosystem and Gemini models. ADK makes it easy to get started with simple agents powered by Gemini models and Google AI tools while providing the control and structure needed for more complex agent architectures and orchestration.<\/p>\n As AI systems evolve from single-purpose tools to collaborative, multi-agent ecosystems, developers need practical guidance on building and orchestrating intelligent agents that can communicate, delegate, and adapt. To bridge this gap, we\u2019ll build a Weather Bot Team<\/b>, a multi-agent system capable of answering weather-related queries while also handling user interactions like greetings, farewells, and safe responses.<\/p>\n This hands-on project aims to demonstrate how to:<\/p>\n By solving this problem, you\u2019ll gain practical experience with ADK\u2019s architecture, orchestration, memory management, and safety best practices, laying the groundwork for more complex, real-world agentic applications.<\/p>\n You can refer to the provided Colab notebook<\/a> to guide you through the hands-on implementation.<\/p>\n Before diving into the code, make sure you\u2019ve completed the following setup steps:<\/p>\n 1. Set-up your Environment & Install ADK<\/strong><\/p>\n Start by creating and activating a virtual environment to isolate your project dependencies:<\/p>\n Now that the environment has been created, we can active it using the following commands:<\/p>\n \u00a0Once your environment is activated, install the Google AI Development Kit (ADK)<\/b>:<\/p>\n 2. Obtain your API Keys<\/strong><\/p>\n You\u2019ll need API keys to interact with different AI models. Grab them from the following sources:<\/p>\n Install the required libraries for the project:<\/p>\n Import libraries:<\/p>\n Set up API Keys:<\/p>\n Define Model Constants for easier use:<\/p>\n In ADK, Tools are the functional building blocks that allow agents to go beyond just generating text. They\u2019re typically simple Python functions that can perform real actions, like fetching weather data, querying a database, or running calculations.<\/p>\n To start, we\u2019ll create a mock weather tool to simulate weather lookups. This helps us focus on the agent\u2019s structure without needing external APIs. Later, we can easily swap it for a real weather service.<\/p>\n Code:<\/b><\/p>\n In ADK, an Agent is the core component that manages the conversation flow, connecting the user, the LLM, and the tools it can use.<\/p>\n To define an agent, you\u2019ll configure a few essential parameters:\u00a0<\/p>\n \u00a0Code<\/b>:<\/p>\n To handle conversations and run the agent effectively, we need two key components:<\/p>\n SessionService<\/b>: This component keeps track of each user\u2019s conversation history and session state. A basic version called InMemorySessionService<\/i> stores all data in memory, making it ideal for testing or lightweight apps. It logs every message exchanged in a session. We\u2019ll dive into saving session data permanently.<\/p>\n Runner<\/b>: This acts as the brain of the system. It manages the entire interaction flow, taking in user input, passing it to the right agent, calling the LLM and any necessary tools, updating session data through the SessionService<\/i>, and producing a stream of events that show what\u2019s happening during the interaction.<\/p>\n \u00a0Code:<\/b>\u00a0<\/p>\nWhat is the Agent Development Kit?<\/h2>\n
Features of Google\u2019s Agent Development Kit<\/h4>\n
\n
Problem Statement<\/b><\/h2>\n
\n
Proposed Workflow<\/h3>\n
![\"Building](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/04\/Soumil-Galat-hai.webp\")
<\/a><\/figure>\n<\/div>\nPrerequisites<\/h2>\n
# Create a virtual environment\n\npython -m venv .venv<\/code><\/pre>\n# Activate the environment\n\n# macOS\/Linux:\n\nsource .venv\/bin\/activate\n\n# Windows CMD:\n\n.venv\\Scripts\\activate.bat\n\n# Windows PowerShell:\n\n.venv\\Scripts\\Activate.ps1<\/code><\/pre>\npip install google-adk<\/code><\/pre>\nSteps to Build Your Weather App<\/h2>\n
Step 1: Setup and Installation<\/h3>\n
# Install Google ADK and LiteLLM\n!pip install google-adk -q\n!pip install litellm -q<\/code><\/pre>\nimport os\n\nimport asyncio\n\nfrom google.adk.agents import Agent\n\nfrom google.adk.models.lite_llm import LiteLlm # For multi-model support\n\nfrom google.adk.sessions import InMemorySessionService\n\nfrom google.adk.runners import Runner\n\nfrom google.genai import types # For creating message Content\/Parts\n\nimport warnings\n\n# Ignore all warnings\n\nwarnings.filterwarnings(\"ignore\")\n\nimport logging\n\nlogging.basicConfig(level=logging.ERROR)<\/code><\/pre>\n# Gemini API Key\n\nos.environ[\"GOOGLE_API_KEY\"] = \"YOUR_GOOGLE_API_KEY\"\n\n# OpenAI API Key\n\nos.environ['OPENAI_API_KEY'] = \u201cYOUR_OPENAI_API_KEY\u201d\n\n# Anthropic API Key\n\nos.environ['ANTHROPIC_API_KEY'] = \u201cYOUR_ANTHROPIC_API_KEY\u201d\n\nprint(\"API Keys Set:\")\n\nprint(f\"Google API Key set: {'Yes' if os.environ.get('GOOGLE_API_KEY') and os.environ['GOOGLE_API_KEY'] != 'YOUR_GOOGLE_API_KEY' else 'No (REPLACE PLACEHOLDER!)'}\")\n\nprint(f\"OpenAI API Key set: {'Yes' if os.environ.get('OPENAI_API_KEY') and os.environ['OPENAI_API_KEY'] != 'YOUR_OPENAI_API_KEY' else 'No (REPLACE PLACEHOLDER!)'}\")\n\nprint(f\"Anthropic API Key set: {'Yes' if os.environ.get('ANTHROPIC_API_KEY') and os.environ['ANTHROPIC_API_KEY'] != 'YOUR_ANTHROPIC_API_KEY' else 'No (REPLACE PLACEHOLDER!)'}\")\n\n# Configure ADK to use API keys directly (not Vertex AI for this multi-model setup)\n\nos.environ[\"GOOGLE_GENAI_USE_VERTEXAI\"] = \"False\"<\/code><\/pre>\nMODEL_GEMINI_2_0_FLASH = \"gemini-2.0-flash\".\n\nMODEL_GPT_4O = \"openai\/gpt-4o\"\n\nMODEL_CLAUDE_SONNET = \"anthropic\/claude-3-sonnet-20240229\"\n\nprint(\"\\nEnvironment configured.\")<\/code><\/pre>\nStep 2: Define Tools<\/h3>\n
def get_weather(city: str) -> dict:\n \"\"\"Retrieves the current weather report for a specified city.\n\n\n Args:\n city (str): The name of the city (e.g., \"Mumbai\",\"Chennai\",\"Delhi\").\n\n\n Returns:\n dict: A dictionary containing the weather information.\n Includes a 'status' key ('success' or 'error').\n If 'success', includes a 'report' key with weather details.\n If 'error', includes an 'error_message' key.\n \"\"\"\n # Best Practice: Log tool execution for easier debugging\n print(f\"--- Tool: get_weather called for city: {city} ---\")\n\n\n city_normalized = city.lower().replace(\" \", \"\") # Basic input normalization\n\n\n mock_weather_db = {\n \"delhi\": {\"status\": \"success\", \"report\": \"The weather in Delhi is sunny with a temperature of 35\u00b0C.\"},\n \"mumbai\": {\"status\": \"success\", \"report\": \"It's humid in Mumbai with a temperature of 30\u00b0C.\"},\n \"bangalore\": {\"status\": \"success\", \"report\": \"Bangalore is experiencing light showers and a temperature of 22\u00b0C.\"},\n \"kolkata\": {\"status\": \"success\", \"report\": \"Kolkata is partly cloudy with a temperature of 29\u00b0C.\"},\n \"chennai\": {\"status\": \"success\", \"report\": \"It's hot and humid in Chennai with a temperature of 33\u00b0C.\"},\n }\n\n\n if city_normalized in mock_weather_db:\n return mock_weather_db[city_normalized]\n else:\n return {\"status\": \"error\", \"error_message\": f\"Sorry, I don't have weather information for '{city}'.\"}\n\n\n# Example usage\nprint(get_weather(\"Mumbai\"))<\/code><\/pre>\nStep 3: Defining the Agent<\/h3>\n
\n
AGENT_MODEL=model\nweather_agent=Agent(\n name=\"weather_agent_v1\",\n model=AGENT_MODEL,\n description=\"Provides weather information for specific cities.\",\n instruction=\"You are a helpful weather assistant. Your primary goal is to provide current weather reports. \"\n \"When the user asks for the weather in a specific city, \"\n \"you MUST use the 'get_weather' tool to find the information. \"\n \"Analyze the tool's response: if the status is 'error', inform the user politely about the error message. \"\n \"If the status is 'success', present the weather 'report' clearly and concisely to the user. \"\n \"Only use the tool when a city is mentioned for a weather request.\",\n tools=[get_weather],\n\n\n)\n\nprint(f\"Agent '{weather_agent.name}' created using model '{AGENT_MODEL}'.\")<\/code><\/pre>\nStep 4: Set up Runner and Session Service<\/h3>\n
# @title Setup Session Service and Runner\n# ---Session Management ---\n# Key Concept: SessionService stores conversation history & state.\n# InMemorySessionService is a simple, non-persistent storage for this tutorial.\n\n\nsession_service=InMemorySessionService()\n\n\n# Define constants for identifying the interaction context\nAPP_NAME=\"weathertutorial_app\"\nUSER_ID=\"user_1\"\nSESSION_ID=\"session_001\"\n\n\n# Create the specific session where the conversation will happen\nsession=session_service.create_session(\n app_name=APP_NAME,\n user_id=USER_ID,\n session_id=SESSION_ID,\n)\n\n\nprint(f\"Session created: App='{APP_NAME}', User=\"{USER_ID}\", Session='{SESSION_ID}'\")\n\n\n# ---Runner ---\n# Key Concept: Runner orchestrates the agent execution loop.\n\n\nrunner=Runner(\n agent=weather_agent,\n app_name=APP_NAME,\n session_service=session_service\n)\n\n\nprint(f\"Runner created for agent '{runner.agent.name}'.\")\n<\/code><\/pre>\nStep 5: Interact with the Agent<\/h3>\n
![\"Building](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/04\/ss.webp\")
<\/figure>\n<\/div>\n![\"Soumil](\"https:\/\/av-eks-lekhak.s3.amazonaws.com\/media\/lekhak-profile-images\/converted_image_reG34wL.webp\")
<\/p>\n