The console keeps data in a chain. Each object belongs to the one above it. The module pages use these words, so read this page once.
Who can do this: everyone. Read it before you start a module.
The objects
Team
A team is the shared space for members, settings, API keys, storage and usage. Everyone in a team sees the same team data, and each person has a team role. See Team settings.
Project
A project holds one kind of recording, robot or human, and everything made from it. Its kind decides how uploads are processed and which checks run. See Projects.
Recording
A recording is one set of data you uploaded: a robot MCAP file, a LeRobot folder or a human first-person recording. Each recording shows its processing status and checks. See Recordings.
Workflow and run
A workflow is the list of processing steps that runs on each upload. A run is one pass of the workflow over one or more recordings. You can rerun a run from a failed step; finished steps keep their results. See Workflows.
Episode
An episode is one task segment cut from a recording. You can review it, label it and choose it for a dataset version. See Episodes.
Dataset version
A dataset version is a set of selected episodes frozen at one moment, with its split, labels and preprocessing. You export it or train on it. See Dataset versions.
Export, training and model
An export packs a dataset version into a format you can download, such as LeRobot. A training job uses a dataset version to produce a model version. A model can be evaluated and deployed. See Export and download, Training and Models.
What you can change
| Object | What you can do |
|---|---|
| Team | Rename it, invite members and change their team roles. |
| Project | Change its name, description and secrets. Owners and Admins manage its members. Owners set the privacy requirement. |
| Recording | Rename it, reprocess it, or move it to the trash. A trashed recording can be restored for 30 days. |
| Episode | Review it, set its task and outcome, and label it. |
| Dataset version | Nothing. A version does not change. Create a new version to use different episodes or settings. |
Dataset versions do not change
Once a dataset version is created, its episodes and settings stay as they were. A training job or an export always refers to the same data. To change the data, create a new version.
Privacy holds
A human recording that needs consent or anonymization is limited until that evidence exists. Until then, raw access is limited to the uploader, the Owners, and the project members who can publish its algorithms and manage its workflow, and exports that include it are blocked. The project’s recording privacy requirement decides whether this rule applies to new uploads.