EmbodiFlowDocs
Get startedDataWorkflows and algorithmsTraining and deploymentDeveloper reference
  • Data

Bring data in

  • Projects
  • Upload recordings
  • Collection orders

Review

  • Playback and visualization
  • Episodes and annotation
  • Quality checks

Datasets

  • Dataset versions
  • Export and download

© 2026 EmbodiFlow

llms.txtOpenAPI JSON
  1. Data
  2. Bring data in
  3. Projects

Projects

Set up a project for one kind of recording, decide who can open it, and keep its settings in shape.

A project is where one stream of data lives, from the first upload to a trained model. It holds the recordings, the workflow that processes them, the episodes, the dataset versions and the models built from them. Most teams create one project per robot setup or per data collection effort.

Open Projects in the sidebar to see every project you can open. Each row shows the kind, the visibility and how many recordings and dataset versions it has. Search by name, slug or ID, filter by Visibility, or sort by Last activity or Name. The ⋯ menu at the end of a row jumps straight to Upload or Settings, or copies the project ID.

Robot and human projects

When you create a project you choose its recording kind, and that choice sticks: it decides which workflow processes your uploads and which checks run. Robot is for teleoperation with arms and grippers, where the robot logged its topics, cameras and joints. Human is for first-person demonstrations recorded with a head-mounted camera rig.

Keep robot and human recordings in separate projects. The kind can't be changed later; the project settings show it as Recording kind (read-only).

Create a project

Only team Owners and Admins can create projects.

1Open New project

Click New project at the top right of the Projects page. It opens a full page, New project, rather than a small dialog.

2Choose the recording kind

Pick Robot or Human. For a robot project, two more fields appear: Robot type and Default task, for example pour beans. Both are optional and can be changed later under Settings; the default task is used for every upload that doesn't name its own task.

3Choose who can open it

Private projects open only for the people you add. Team projects open for everyone on the team.

4Name it and create

Enter a Project name and an optional Description, then click Create. The project opens on its upload page, ready for your first recordings.

Privacy is required by default, so recordings are anonymized before others can read or export them. If no people ever appear in the data, a team Owner can change that in the project settings.

Members and project roles

In a private project, members are how people get in. Open the project, choose Settings in the sidebar, then Members. You can add an existing teammate or type an email under Invite member and click Send invitation; the person joins with the roles you picked once they accept. A member can hold several roles and gets everything each one allows.

RoleWhat it allows
OwnerFull control of this project.
EditorChange data and start processing. Cannot manage members or the project.
ReviewerReview annotations, quality and approvals.
LabelerLabel the items assigned to them.
ViewerRead and download data; cannot change anything or start processing.
UploaderUpload data and start processing.
CollectorUpload to the collection orders assigned to them.

To change someone's access later, open the ⋯ menu on their row and choose Manage access, or Remove member to take them off the project. Team Owners, team Admins and project Owners manage members; everyone else sees the list read-only. The team-wide roles are explained in Roles and permissions.

Project settings

Settings sits at the bottom of the project sidebar menu. It has five sections on the left.

  • General: the project name, description, robot type and default task. Click Save changes when you're done.
  • Members: who is on the project and with which roles, as described above.
  • Annotation schema: the label fields annotators use (Task, Outcome, Relevance, Subtasks, Subtask outcome), the label languages, whether human edits need a reviewer, whether episodes may overlap, and your subtask labels and attributes. Click Save schema to keep the changes.
  • Secrets: values a workflow step can read as environment variables, such as an access key, without writing them into the workflow. A value is write-only: once saved you can replace it but never read it back.
  • Trash: the deleted items of this project.

General also holds the Recording privacy requirement. Privacy required means recordings need anonymization before anyone else can read or export them. Privacy not required is only for data with no people or personal content, and asks you to write down why; the change goes into the audit log. Only a team Owner can change it. It applies to new uploads, and you can choose to apply it to the existing recordings too; processing results and dataset versions are never rewritten.

At the bottom of General is the Danger zone. Archive project keeps the project visible but read-only. Delete project moves it to the trash.

Trash and restore

Deleted projects, recordings, episodes, dataset versions and models wait in the trash for 30 days. During that time you can bring any of them back with Restore; a restored project comes back with its items. Each row shows the date until which it can be restored.

The team trash is in the team menu: click the team name at the bottom of the sidebar, then Trash. A project's own trash is under Settings › Trash.

Delete permanently can't be undone: you type the item's name to confirm, and it's gone. Storage is billed until items are permanently deleted.

Next

  • Upload recordings
  • Workflows
  • Roles and permissions
Previous

Data

Next

Upload recordings