Google ADK Testing & Evaluation

AgentCI delivers enterprise-grade Google ADK testing and evaluation for teams building with Google's Agent Development Kit. Our automated Google ADK CI platform discovers Agent and LlmAgent instances, validates multi-agent systems, and runs comprehensive evaluations - with zero configuration required.

Google ADK testing for Gemini-powered agent applications

AgentCI automatically discovers and evaluates Google ADK agents, including:

  • Agent discovery: Agent(), LlmAgent(), and multi-agent systems with sub_agents
  • Evaluation types: Accuracy, safety, performance, and coordination testing for Gemini models
  • CI/CD integration: Automated testing on pull requests via GitHub
  • Zero code changes: No annotations or modifications to your Python code required

Supported Agent Patterns

Agent Class

from google.adk.agents import Agent

agent = Agent(
    name="weather_assistant",
    model="gemini-2.0-flash",
    description="Provides weather information for cities.",
    instruction="You are a helpful weather assistant.",
    tools=[get_weather, get_current_time]
)

LlmAgent with Sub-Agents

from google.adk.agents.llm_agent import LlmAgent

coordinator_agent = LlmAgent(
    name="coordinator",
    model="gemini-2.5-flash",
    description="Coordinates multiple specialized agents.",
    instruction="You delegate tasks to specialized sub-agents.",
    sub_agents=[weather_agent, research_agent]
)

Supported Tool Patterns

Plain Functions

def get_weather(city: str) -> dict:
    """Retrieves the current weather report for a specified city.

    Args:
        city (str): The name of the city.

    Returns:
        dict: Weather report with status and result.
    """
    return {
        "status": "success",
        "report": f"The weather in {city} is sunny, 72°F."
    }

What Gets Auto-Discovered

AgentCI automatically finds:

  • Agent() and LlmAgent() instances with name, model, instruction, and tools parameters
  • Multi-agent systems with sub_agents
  • Plain Python functions with docstrings and type hints
  • Tools wrapped with FunctionTool

No decorators required - Google ADK's function-based approach works seamlessly with AgentCI.

Next Steps