18-callbacks

18. Callbacks

Intercepting and customizing agent behavior.

Blog Post: https://arjunprabhulal.com/adk-callbacks/

Table of Contents

1. Overview
2. Prerequisites
3. Setup Steps
4. Callback Types
5. Running the Demo
6. Next Steps

Overview

Callbacks let you intercept and customize agent behavior at various points in the execution lifecycle:

• Logging - Track agent activity
• Filtering - Block unwanted content
• Validation - Ensure data quality
• Rate Limiting - Control API usage

Prerequisites

• Python 3.10+
• Gemini API key from AI Studio

Setup Steps

1. Navigate to this module:

cd 18-callbacks

2. Create and activate a virtual environment:

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

pip install -r ../requirements.txt

4. Set up environment variables in logged_agent/.env:

GOOGLE_API_KEY=your-api-key-here

Callback Types

Callback	When It Runs	Use Case
before_agent_callback	Before agent starts	Setup, validation
after_agent_callback	After agent completes	Cleanup, logging
before_model_callback	Before LLM call	Prompt modification
after_model_callback	After LLM response	Response filtering
before_tool_callback	Before tool execution	Input validation
after_tool_callback	After tool result	Output processing

Example

def before_model_logging(callback_context, llm_request):
    print(f"Sending request to model...")
    return None  # Continue normally

agent = Agent(
    before_model_callback=before_model_logging,
)

Running the Demo

cd logged_agent
python agent.py

The demo shows:

• Logging callbacks for all lifecycle events
• Request/response inspection
• Execution timing

Next Steps

Continue to 19. Artifacts