Concepts

OneOpen ML Studio uses a clear platform hierarchy. Workspaces and projects are not the same thing.

Platform hierarchy

Platform
└── Organization
    ├── Workspaces
    │   ├── Projects
    │   │   ├── Datasets
    │   │   ├── Annotation tasks
    │   │   ├── Pipelines
    │   │   ├── Training runs
    │   │   └── Models
    │   └── Members
    └── Organization settings

Level	Meaning
Organization	Tenant / company / lab boundary (billing, SSO, quotas)
Workspace	Team area inside an org (members, shared projects)
Project	One ML use case — owns data, labels, versions, train, models

Local mode auto-creates a Personal organization with a Default workspace. New projects attach there unless you pick another workspace.

Today

Default URL: http://127.0.0.1:8765
Persistence: SQLite + files under ~/.oneopen (Postgres / Redis / MinIO in Team via Compose)
Hierarchy pages: /orgs, /orgs/{id}, /workspaces/{id}
Project UI: annotate, quality/clean, versions, train, pipelines, experiments
Team/Enterprise: ONEOPEN_DEPLOYMENT_MODE + /setup for auth

Projects

A project is one task (e.g. PPE detection or RAG eval). It owns classes/samples, annotations, dataset versions, pipelines, training runs, and registered models.

Classes, images, annotations (visual)

Classes are named labels with colors (match model class names for Label Assist).
CV annotations: bbox / polygon / whole-image classification with sources manual | yolo | sam | import.
Non-image projects label samples (spans, values, transcripts, LLM fields) instead of canvas geometry.

Splits and versions

Images are train / valid / test / unassigned. A dataset version records split ratios, preprocess, and augmentation used for export or training.

Architecture overview: Architecture.
Roadmap: Roadmap.