On this page
PydanticAI extends the Pydantic ecosystem with type-safe AI agents that validate structured outputs and integrate tools. Running them with ZenML puts these agents into production-ready pipelines, ensuring reproducibility, artifact lineage, and deployment flexibility.
Features with ZenML
- Type-safe orchestration. Run PydanticAI agents inside ZenML pipelines with guaranteed structured outputs.
- Artifact tracking. Store queries, validated responses, and error cases as ZenML artifacts.
- Error handling. Integrate post-run checks or evaluation steps to catch invalid responses.
- Composable workflows. Combine PydanticAI with retrieval, evals, and deployment steps in a single pipeline.
- Portable execution. Move seamlessly from local runs to scalable orchestrators like Kubernetes or Airflow.
Main Features
- Type-safe agents. Ensure AI outputs conform to strict schemas defined with Pydantic.
- Simple run_sync API. Call agents with a straightforward synchronous interface.
- Tool integration. Extend agents with callable Python functions as tools.
How to use ZenML with PydanticAI
from zenml import ExternalArtifact, pipeline, step
from pydanticai_agent import agent
@step
def run_pydanticai(query: str) -> str:
result = agent.run_sync(query)
return str(result.output)
@pipeline
def pydanticai_agent_pipeline() -> str:
q = ExternalArtifact(value="What is the secret data?")
return run_pydanticai(q.value)
if __name__ == "__main__":
print(pydanticai_agent_pipeline())
