Google ADK Agent

Google ADK lets you build Gemini-powered agents with a simple callable interface and built-in tools; connecting ADK to ZenML wraps those agents in reproducible pipelines with artifact tracking, observability, and an easy path from local experiments to production.

Features with ZenML

• Pipeline orchestration. Run Google ADK agents as ZenML steps inside reproducible pipelines.
• ‍Artifact management. Capture agent inputs and outputs for lineage, versioning, and auditability.
• ‍Evaluation ready. Add post-run checks or eval steps to monitor quality over time.
• ‍Infrastructure agnostic. Scale from local runs to Kubernetes, Airflow, and other ZenML stacks.
• ‍Composable workflows. Combine agents with retrieval, evals, and deployment steps in one DAG.

Main Features

• Gemini-powered agents. Build on Google’s latest models through ADK.
• ‍Simple callable interface. Invoke agents directly or through common run and execute methods.
• ‍Built-in tools. Integrate calculations and custom tools inside agent workflows.

How to use ZenML with Google ADK Agent

from zenml import ExternalArtifact, pipeline, step
from adk_agent import root_agent

@step
def run_adk(query: str) -> str:
    return str(root_agent.run(query))


@pipeline
def google_adk_pipeline() -> str:
    q = ExternalArtifact(value="What's the weather like in Tokyo?")
    return run_adk(q.value)

if __name__ == "__main__":
    print(google_adk_pipeline())

Additional Resources

• Google ADK Documentation