SNA-Agentic-AI-Examples

A repository demonstrating examples of Agentic AI using the Google Agent Developer Kit (ADK).

Getting Started

Follow these steps to set up your environment and run the Agentic AI examples.

Prerequisites

Python 3.11 installed on your system.

A Google API Key for access to the Gemini API (this is not needed you use Datalab LLM)

API Key Setup

Obtain a Key: Get a free Google API Key from the Google AI Studio: https://aistudio.google.com.

Create .env file: In the root directory of this repository, create a file named .env.

Add Key: Inside the .env file, add your API key with the following format:
Bash

GOOGLE_API_KEY="YOUR_API_KEY_HERE"

Using Alternative LLMs (Datalab)

if you want to use alternative LLMs from the Datalab server at Aristotle University, add these additional variables to your .env file:

Note: You must be connected to the AUTH VPN to access the Datalab's LLM.

OLLAMA_API_BASE="https://babili.csd.auth.gr:11435"

Environment Setup

Install Dependencies: Open your terminal in the root directory and install the required packages:
Bash

pip install -r requirements.txt

Running the Examples

This repository includes multiple methods for running the agents: the ADK Web UI, Command-Line Runners, and Docker-based persistent sessions.

Examples 1-4: Using the ADK Web UI

The Google ADK provides a local web interface for easily testing and interacting with your agents.

1. Start the Local Server: From the root directory, run the following command:

   adk web

2. Access the UI: A local server will start, typically accessible at: http://localhost:8000

3. Select Examples: You can select and run the first four examples directly from the ADK Web UI environment.

Example Details:

• 01_hello_world_agent: Basic agent setup and interaction
• 02_tool_calling_agent: Agent with tool integration
• 03_agent_as_tool_calling_agent: Hierarchical agent structure where one agent uses another as a tool
• 04_output_schema_tool_calling_agent: Agent with structured output schemas

Example 5: Command-Line Agent Runner (In-Memory Sessions)

The fifth example demonstrates how to run an agent directly from the command line with in-memory session management.

Run the Agent: Execute the Python script from your terminal:

python 05_sdk_agents_operations/agent_runner.py

The agent will execute its defined workflow and output the results directly in the terminal using in-memory session storage.

Example 6: Command-Line Agent Runner with Database Sessions

This example demonstrates persistent session management using PostgreSQL for storing conversation history.

Setup PostgreSQL with Docker

1. Build the Docker image:

   docker build -t postgres-agent -f 06_sdk_agents_operations/Dockerfile 06_sdk_agents_operations/

2. Run the PostgreSQL container:

   docker run -d -p 5432:5432 --name postgres-agent postgres-agent

   This creates a PostgreSQL database with:

   • Username: agent_user
   • Password: agent_password
   • Database: agent_sessions
   • Port: 5432

3. Run the Agent:

   python 06_sdk_agents_operations/agent_runner.py

Key Features:

• Persistent session storage in PostgreSQL
• Conversation history maintained across runs
• Session management with user and application context
• Same user can continue conversations by reusing session IDs

Managing the Docker Container:

• Stop container: docker stop postgres-agent
• Start container: docker start postgres-agent
• Remove container: docker rm postgres-agent
• View logs: docker logs postgres-agent

Example 7: Agent with Callbacks

This example demonstrates using callbacks to save session data after each agent interaction, running through the ADK Web UI.

Run with ADK Web UI:

adk web

Then select example 07_agent_callback from the UI.

Key Features:

• Implements after_agent_callback to execute custom logic after each agent turn
• Automatically saves complete session history to JSON files
• Sessions saved to saved_sessions/ directory
• Each session file named as session_{session_id}.json
• Useful for debugging, logging, and conversation analysis

Note: The saved_sessions/ directory will be created automatically if it doesn't exist.