Google Agent Development Kit (ADK) Crash Course

Learn how to add built-in tools to Google Agent Development Kit agents, including the Google search tool and code execution, and run a tool-enabled agent to fetch internet results.

Explore custom tools in the adk, focusing on function tools, docstrings, and structured returns, with examples like get current time and wrapping tools as agents.

Identify the limitation of Google agent development kit: you cannot list a custom tool with a built-in tool, as Adcc can error, requiring removal of the custom tool.

Connect your sdk agents to multiple LLM providers with Lite LLM, enabling OpenAI and Claude access alongside Google Gemini, using simple installation, API keys, and practical demos.

Define structured outputs with an output schema and a Pydantic model to ensure agents return json with a subject and body, stored under an output key for cross-agent sharing.

Pull back the curtain on core adk components, including session management, memory management, events, and runners, to enable deploying sophisticated agents in real-world settings beyond the sdk web.

Orchestrate the adc memory management workflow by detailing sessions, session identifiers, state, and events, with a runner selecting the right agent to process requests, perform tool calls, and return responses.

Examine the three session types in the ADK: in-memory sessions lose data on restart, Vertex AI sessions store in Google Cloud, and persistent database sessions survive restarts.

See a coding example where sessions, state, and runners unite to chat with an agent. Learn about in-memory sessions, environment setup, and a qa agent.

Introduce persistent agent sessions by using ADC's database session service to store context in a SQLite database, ensuring continuity across restarts and a consistent user experience.

Set up a reminder agent with a persistent SQLite database, initialize sessions, and run an interactive loop to add and query reminders using the adk runner and async agent calls.

Discover how the memory agent uses the 2.0 flash model to manage a username and reminders, with tools to add, view, delete reminders, updating and persisting the session state.

Explore how the ADK generates a my agent db file, with sessions, events, and reminders, and how tool calls and messages are persisted to memory or a database.

Explore how the ADK uses a root manager to delegate requests to sub agents, such as a crypto analyst or a book recommender, each with tools like crypto price tool.

Learn to use built-in tools like Google search in the ADK by wrapping them as agent tools, since route agents delegate to subagents who use only custom tools.

Build a manager agent that delegates user requests to crypto analysts and a book recommender, using tools for news and current time as fallbacks when needed.

Run the sdk web server from the multi agents folder, select the manager agent, and see requests delegate to sub agents like book recommender and crypto analyst.

Understand a multi-agent stock trading system with a shared state, where a stock assistant coordinates policy, trade, research, and order agents to manage a user portfolio.

Navigate main.py in a multi-agent system, covering session management (in-memory or database), an interactive user loop, and subagents for history and agent calls.

Explore how the root stock assistant delegates tasks to policy, trade, research, and orders agents, manages user state and portfolio, and updates interaction history for retail investor questions.

Explore the stock research agent and order agent, which use a shared state to review holdings, simulate selling, and update portfolios with timestamped interaction history.

Run the multi-agent system demo to simulate buying and selling stocks and diversify your demo portfolio with the stock, trade, portfolio orders, and research agents.

Learn six types of callbacks in the Google agent development kit crash course, including before and after agent, model, and tool callbacks, with examples of logging and pre- and post-processing.

Run the code in the google adk, observe before and after agent callbacks, and track ticket state and processing time to update ticket counts and analytics.

Use the before model callback to pre-check the user's last message for profanity or spam before sending the LM request to models like OpenAI, Gemini, or Claude.

Learn to use the after model callback to format responses by appending a footer with timestamp, duration, and agent name, and to detect words for replacement.

Explore the before tool callback and after tool callback, showing how pre-call validation, authorization checks, and autocorrection refine tool arguments for stock quotes like AAPL while blocking restricted tickers.

Understand the after two callback to format price and percent change to two decimals, and trigger a volatility alert when moves exceed five percent, demonstrated with Tesla and Nvidia quotes.

Explore sequential workflow agents that execute subagents in an order, illustrated by a mortgage loan pipeline with a loan validated agent, risk scorer agent, and loan recommender agent, via state.

Run the sequential agent in Adcc web to validate loan data and score risk. Observe how complete, missing, and high-risk inputs determine approval or decline.