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.

From Python

Use flyte.remote.Artifact.create(). It 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)

If you do not pass a version, one is generated for you. Pass attrs, kind, and card the same way as in task-produced metadata. The call is synchronous by default; use Artifact.create.aio(...) from async code.

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, external_ref is the provenance you can attach.

From the CLI

flyte create artifact publishes a file directly:

flyte create artifact my-model --from-file model.pt --kind model --attr framework=torch
flyte create artifact llama3 --from-file weights.bin --external-ref hf://meta-llama/Meta-Llama-3-8B
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.

Finding artifacts

flyte get artifact browses the registry:

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

The same queries are available in Python. Artifact.get(name) returns the latest version by default, Artifact.listall() iterates versions newest first with server-side filtering, and Artifact.list_names() lists distinct names with their version counts.

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.