CabaModel: Gemini-Native Agent Architecture

An advanced study on AI agent architecture using Google ADK (Agent Development Kit), FastAPI, Pydantic v2, and asynchronous event-driven execution.

This project demonstrates a production-ready transition from simple scripts to a resilient, microservice-oriented agent orchestration.

🚀 Key Features

• Google ADK Runner Architecture: Implements the latest Runner pattern for managing agentic loops, event streams, and tool calls.
• RESTful Interface: Powered by FastAPI with an interactive Swagger UI for real-time testing and orchestration.
• Hexagonal Design: Clear separation between Domain, Application, and Infrastructure layers, ensuring scalability and testability.
• Asynchronous & Non-blocking: Full asyncio implementation to handle I/O and event streams without blocking the main event loop.
• Strict Validation: Pydantic v2 ensures data integrity across all agent-tool interactions.
• Resilience: Built-in retry logic with exponential backoff for robust API communication.

📂 Project Structure

• src/cabamodel/domain/ – Core models and Pydantic schemas.
• src/cabamodel/infrastructure/ – API implementation, ADK Adapters, and Tools.
• src/cabamodel/application/ – Specialized agent definitions (e.g., temporal_agent, c4b4_bot).
• main.py – Entry point for the FastAPI server and agent initialization.

⚙️ How to Run

1. Clone the repository:

   git clone https://github.com/your-username/CabaModel.git

2. Setup Environment: Create a .env file based on .env.example and add your GOOGLE_API_KEY.

3. Install Dependencies:

   python -m venv venv
   .\venv\Scripts\activate
   pip install -e .

4. Run the API:

   python main.py

5. Interact: Open your browser at http://127.0.0.1:8000/docs to access the interactive Swagger documentation.

🛠️ Programmatic Usage Example

from src.cabamodel.application.temporal_agent import root_agent
from src.cabamodel.infrastructure.agent_service import run_agent_async

async def main():
    # Execute the agent using the ADK Runner-based service
    response = await run_agent_async(root_agent, "What time is it now?")
    print(f"Agent Response: {response}")

# This pattern handles tool-calling loops and event streams automatically

⚖️ License

Distributed under the Apache 2.0 License. See LICENSE for more information.