Integrations

Inferpathio integrates with the orchestration tools, cloud ML platforms, and CI/CD systems your team already uses. Each integration is configured via the SDK or a small YAML block in your pipeline definition.

Apache Airflow

Use the IFPTrackOperator to wrap any training task with automatic experiment tracking and model registration.

from inferpathio.integrations.airflow import IFPTrackOperator

train_task = IFPTrackOperator(
    task_id="train_fraud_model",
    model_name="fraud-detector",
    python_callable=train_fn,
    dag=dag
)

Kubeflow Pipelines

Add IFP tracking to any Kubeflow component with a single decorator:

from inferpathio.integrations.kubeflow import ifp_component

@ifp_component(model_name="churn-predictor")
def train_component(data_path: str, output_model: str):
    # your training code
    ...

GitHub Actions

Use the official Action to post experiment results as PR comments and gate promotion:

- name: Track with Inferpathio
  uses: inferpathio/action@v2
  with:
    api_key: ${{ secrets.IFP_API_KEY }}
    model_name: my-model
    script: python train.py

AWS SageMaker

Pass Inferpathio as a callback during estimator fitting:

from inferpathio.integrations.sagemaker import IFPCallback

estimator.fit(
    inputs=data_channels,
    callbacks=[IFPCallback(model_name="sage-classifier")]
)

Google Vertex AI

Use the Vertex AI custom training job adapter:

from inferpathio.integrations.vertexai import IFPVertexTracker

tracker = IFPVertexTracker(
    model_name="vertex-nlp",
    project="my-gcp-project"
)
tracker.attach_to_job(custom_job)

Environment variables

All integrations respect the following environment variables, useful in containerized environments:

  • IFP_API_KEY — your API key (required if not using ifp configure)
  • IFP_API_BASE — override the API base URL (default: https://api.inferpathio.com/v1)
  • IFP_DISABLE_TRACKING — set to 1 to run in dry-run mode without logging