Quick start
Get from zero to a running local ML Studio in a few minutes.
1. Install
pip install "oneopen-ml-studio[yolo]"
2. Initialize data directory
oneopen init
Creates ~/.oneopen (or ONEOPEN_DATA_DIR) with SQLite and folders for projects, exports, and models.
3. Start the server
oneopen start
Open http://127.0.0.1:8765.
(oneopen serve is the same command.)
Important
Local mode (default) binds 127.0.0.1 and does not require login. For Team/Enterprise, set ONEOPEN_DEPLOYMENT_MODE=team (or enterprise), open http://127.0.0.1:8765/setup, then use sessions or API tokens. Do not expose Local mode on untrusted networks.
Useful options
oneopen start --port 9000
oneopen start --host 127.0.0.1 --reload
4. Create a project
New project (or
/projects/new).Name + type — object detection is the recommended first path (classification and segmentation also work).
Optional custom location; blank →
~/.oneopen/projects/<id>/.
5. Add classes and images
Add at least one class.
Upload images or import a YOLO folder/zip.
6. Annotate (visual)
Open Annotate.
Draw boxes or polygons (or assign a whole-image class for classification projects).
Optionally run Label Assist (YOLO) or SAM point prompts, then correct.
7. Version, train, or export
Assign splits.
Create a dataset version.
Export and/or Train locally (Ultralytics / LibreYOLO / Torchvision).
Typical first workflow (CV)
Create project → Add classes → Upload images
↓
Annotate (manual + Label Assist / SAM)
↓
Assign splits → Generate version
↓
Export YOLO/COCO and/or Train locally
Next steps
Concepts — today vs target platform
Product vision — full product vision
Training — training jobs
Data prep & training by task (with screenshots) — screenshot guide per modality
Configuration — environment variables