Using the Python API

While vertex-deployer is primarily a CLI tool, you can also use the VertexPipelineDeployer class directly in your Python code. This is useful for custom scripts, notebooks, or when you need more control over the deployment process.

Basic example

from pathlib import Path
from deployer.pipeline_deployer import VertexPipelineDeployer
from my_project.vertex.pipelines.my_pipeline import my_pipeline

deployer = VertexPipelineDeployer(
    pipeline_name="my_pipeline",
    pipeline_func=my_pipeline,
    project_id="my-gcp-project",
    region="europe-west1",
    staging_bucket_name="my-staging-bucket",
    service_account="my-sa@my-gcp-project.iam.gserviceaccount.com",
    gar_location="europe-west1",
    gar_repo_id="my-pipelines-repo",
    local_package_path=Path("."),
)

# Compile, upload, and run in one call
deployer.compile_upload_run(
    enable_caching=True,
    parameter_values={"learning_rate": 0.01, "epochs": 10},
    tags=["latest"],
)

Constructor parameters

Parameter	Type	Required	Description
`pipeline_name`	`str`	Yes	Name of the pipeline (must match the `@kfp.dsl.pipeline` function name)
`pipeline_func`	`Callable`	Yes	The kfp pipeline function decorated with `@kfp.dsl.pipeline`
`run_name`	`str`	No	Custom run name displayed in the UI. Defaults to `{pipeline_name}-{tag}-{timestamp}`
`project_id`	`str`	No	GCP project ID
`region`	`str`	No	GCP region (e.g. `europe-west1`)
`staging_bucket_name`	`str`	No	GCS bucket name for pipeline staging (without `gs://` prefix)
`service_account`	`str`	No	Service account email for pipeline execution
`gar_location`	`str`	No	Google Artifact Registry location
`gar_repo_id`	`str`	No	Google Artifact Registry repository ID (KFP format)
`local_package_path`	`Path`	No	Local directory for compiled pipeline YAML files

Step-by-step usage

Each method returns self, so you can chain calls:

# Step by step
deployer.compile()
deployer.upload_to_registry(tags=["latest", "v1.0"])
deployer.run(
    enable_caching=True,
    parameter_values={"model_name": "my-model"},
    experiment_name="my-experiment",
    tag="latest",
)

# Or chained
deployer.compile().upload_to_registry(tags=["latest"]).run(
    parameter_values={"model_name": "my-model"},
    tag="latest",
)

Methods

`compile()`

Compiles the pipeline function to a YAML file in the local_package_path directory using kfp.compiler.Compiler.

`upload_to_registry(tags=["latest"])`

Uploads the compiled pipeline YAML to Google Artifact Registry. Requires gar_location and gar_repo_id to be set.

`run(enable_caching, parameter_values, input_artifacts, experiment_name, tag)`

Submits a pipeline run to Vertex AI. The pipeline template is resolved from Artifact Registry if available, otherwise from the local compiled YAML.

enable_caching: None uses compile-time defaults. True/False overrides caching for all tasks.
parameter_values: Dict of pipeline parameter names to values.
input_artifacts: Dict of artifact parameter names to resource IDs (e.g. {"vertex_model": "456"}).
experiment_name: Groups runs under a Vertex Experiment. Defaults to {pipeline_name}-experiment.
tag: Tag of the uploaded pipeline version to use from Artifact Registry.

`compile_upload_run(enable_caching, parameter_values, experiment_name, tags)`

Convenience method that calls compile(), optionally upload_to_registry(), and run() in sequence.

`schedule(cron, enable_caching, parameter_values, tag, delete_last_schedule, scheduler_timezone)`

Creates a recurring schedule for the pipeline on Vertex AI. See Scheduling Pipelines for details.

Running without Artifact Registry

If you don't need Artifact Registry, omit gar_location and gar_repo_id. The deployer will use the locally compiled YAML file instead:

from pathlib import Path

deployer = VertexPipelineDeployer(
    pipeline_name="my_pipeline",
    pipeline_func=my_pipeline,
    project_id="my-gcp-project",
    region="europe-west1",
    staging_bucket_name="my-staging-bucket",
    service_account="my-sa@my-gcp-project.iam.gserviceaccount.com",
    local_package_path=Path("."),
)

deployer.compile().run(parameter_values={"learning_rate": 0.01})

Full API reference

For the complete API reference with all method signatures, see the auto-generated API docs.