Publish your own artifacts

Not every dataset or model is produced by a task. You can publish an existing asset as an artifact from anywhere: a laptop, a notebook, a CI job, or an external pipeline. This is how you bootstrap a registry from assets you already have.

Publish an artifact

Programmatic

CLI

flyte.remote.Artifact.create() uploads the value and registers a version in one call:

        
    
import flyte
from flyte.io import File
from flyte.remote import Artifact

flyte.init_from_config()

published = Artifact.create(
    File.from_local_sync("data/2026-08-18.csv"),
    name="incoming_dataset",
    description="Dataset dropped off by a partner",
    external_ref="s3://partner-bucket/drop/2026-08-18.csv",
)
print(published.name, published.version)

The call is synchronous by default; use Artifact.create.aio(...) from async code.

flyte create artifact publishes a file directly:

        
    
flyte create artifact incoming_dataset --from-file data/2026-08-18.csv \
    --external-ref s3://partner-bucket/drop/2026-08-18.csv
flyte create artifact my-model --from-file model.pt --kind model --attr framework=torch
flyte create artifact my-model --from-file model.pt --card model_card.html --card-type model

--attr is repeatable, --kind is one of model, data, or generic, and the card format is inferred from the file extension.

If you do not pass a version, one is generated for you. Set attrs, kind, and a card the same way as in task-produced metadata.

The external ref records where the data came from. When Artifact.create() runs inside a task, the producing run is stamped on the artifact automatically; outside a task, the external ref is the provenance you can attach.

Finding artifacts

Programmatic

CLI

        
    
from flyte.remote import Artifact

Artifact.list_names()                                   # all artifact names, with latest version info
Artifact.listall("my-model")                            # every version of my-model, newest first
Artifact.get("my-model", version="1.0")                 # one version
Artifact.list_names(search="model")                     # names containing "model"
Artifact.listall(source_run="my_run")                   # versions produced by a run
Artifact.listall(kind="model", attrs={"framework": "torch"})

        
    
flyte get artifact                        # all artifact names, with latest version info
flyte get artifact my-model               # every version of my-model, newest first
flyte get artifact my-model 1.0           # details of one version
flyte get artifact --search model         # names containing "model"
flyte get artifact --source-run my_run    # versions produced by a run
flyte get artifact --kind model --attr framework=torch

Artifact.get(name) returns the latest version by default, and listall and list_names return iterators. For partitions, ranges, and more filters, see Find and retrieve artifacts.

Artifacts are scoped to a project and domain. All of these calls accept project and domain arguments, and default to the ones in your config.