Upload. Version. Trace.
Three tight primitives. One reproducible data workflow.
01
Upload
Upload MCAP, Parquet, Zarr, or LeRobot data and run auditing to get metrics.
MCAPParquetZarrLeRobotCustom
02
Version
Commit dataset snapshots with history that your team can actually revisit.
03
Trace
Add a few lines to your training script to link every run back to its dataset.
run = hephaes.start_run(
dataset="snapshot_v24",
model="pi0-fast")
Pricing
Free to start. Built for teams.
Best for
Solo evaluation and prototyping
Shared dataset workflows across a robotics team
Repos
1 repo
More repos for active team projects
Storage
Enough to test and explore
Scales with production usage
Users
1 user
Multiple collaborators
Conversion usage
Monthly included credits
Larger included pool + additional usage
Shared configs
Personal configs
Team-wide reusable configs
Collaboration
Shared repos, permissions, reviewable workflows
Auditability
Basic history
Full team audit trail