SDK
Python SDK
The Python SDK is the primary developer surface for working with platform entities inside notebooks, scripts, CI jobs, and internal services. It is designed for teams that want full lifecycle automation without reducing the platform to raw HTTP calls or manual UI interaction.
What The Python SDK Should Cover
Data and retrieval
- query datasets, episodes, annotations, and benchmark packs
- search by metadata and semantic criteria
- export manifests and dataset slices without losing version identity
Workflow actions
- trigger curation, finalization, training, evaluation, release, and deployment workflows
- poll or subscribe to workflow state
- inspect failure context programmatically
Training and model lifecycle
- launch runs against immutable dataset snapshots
- inspect experiments, checkpoints, and candidate models
- read release evidence and post-deploy rollout state
Operational queries
- inspect fleets, deployments, incidents, maintenance cases, and intervention summaries
- build internal dashboards or automation without scraping the UI
Example
from rfabric import RFabricClient
client = RFabricClient(api_key="rf_xxx")
dataset = client.datasets.get("folding_v12")
episodes = dataset.search(
query="failed grasp on shirt corner",
score_threshold=0.82,
site="staging_eu"
)
pack = client.release_packs.get("folding_release_v5")
run = client.training.launch(
dataset=dataset.id,
config="configs/folding_sweep.yaml",
evaluation_pack=pack.id,
)Why Teams Care
Research fit
Teams can stay in Python for analysis, experimentation, and automation.
Lifecycle continuity
Programmatic work still operates on governed platform objects.
Faster internal tooling
Teams can build higher-level automation and dashboards without inventing separate state models.