Workspaces are a similar concept to kubernetes namespaces since their main intention is to provide and shared environment across multiple users (or teams). Unlike kubernetes, kaos absolutely requires a workspace for all work. A workspace is ideally an environment for training a single "type" of model - for example mnist.

A workspace consists of only a single dynamic train pipeline - a requirement to have full provenance while incrementally developing machine learning models.

Workspaces can be shared by connecting to an existing workspace via kaos workspace set assuming the workspace has been first created (i.e. kaos workspace create).

It is not necessary to use multiple workspaces for a slightly different model!