Datasets & Apps

Datasets

OSMO datasets are content-addressable, deduplicated, versioned bundles of files produced by workflows.

from strands_osmo import osmo_dataset_list, osmo_dataset_describe

osmo_dataset_list(tag="latest")
osmo_dataset_describe(name="thor-benchmark", version="2026-05-14-001")

To produce a dataset, declare it in your workflow's outputs:

tasks:
  - name: train
    outputs:
      - dataset:
          name: my-checkpoint
          tags: [latest, prod]

Apps

An app is a reusable workflow spec - a versioned template you submit later with different inputs.

from strands_osmo import osmo_app_create, osmo_app_list

osmo_app_create("my-pipeline.yaml", name="train-imagenet", version="1.0.0")
osmo_app_list(name="train-imagenet")