Developer Tools

rFabric CLI

The CLI is the terminal-native surface for the full lifecycle. It allows engineers and operators to inspect datasets, trigger workflows, launch training, review release state, and manage fleets without reducing the platform to UI clicks or one-off scripts.

What The CLI Should Expose

Data commands

ingest or push sessions
inspect datasets and manifests
search or export curated snapshots
inspect annotation and curation state

Training commands

submit runs
compare run and experiment state
inspect checkpoints and candidate models

Release commands

attach benchmark packs
review promotion status
approve or reject candidates
inspect rollout eligibility

Fleet and ops commands

inspect deployment state
target cohorts
monitor update progress
inspect incidents, maintenance, and intervention summaries

Example Session

$ rfabric auth login
Successfully authenticated as admin@company.com

$ rfabric data push ./local_lerobot_dir --dataset folding_v12
Uploaded 124 episodes to folding_v12

$ rfabric train submit --config sweep.yaml
Job launched: run_8819

$ rfabric release promote run_8819 --pack folding_release_v5
Candidate approved for staging rollout

$ rfabric fleet deploy model:folding_v5 --target staging_fold_arms
Canary rollout started

Why Teams Care

Automation

CI jobs and scripts can operate on first-class platform objects.

Operational speed

When something is going wrong, terminal access is often the fastest path to inspection and action.

Consistency

The CLI should use the same resource names and lifecycle transitions as the UI, SDKs, and API.