oneopen-ml-studio 1.0.0

Open-source local-first ML data platform: annotate, version, and train CV datasets — expanding to multi-modal collaborative ML workflows

OneOpen ML Studio

Open-source, local-first ML data platform
From raw data → annotation → validated versions → training-ready datasets → models.

---

Product definition

An open-source, local-first and self-hosted collaborative machine-learning data platform for preparing, annotating, validating, versioning, training and managing ML datasets and models.

Core workflow (target platform):

Raw Data → Profiling → Annotation → Review → Quality Validation
→ Transformation → Dataset Version → Training Readiness
→ Model Training → Experiment Comparison → Model Registry → Active Learning

Differentiation:

Multi-modal annotation
+ Training-ready dataset generation
+ Reproducible versioning
+ Multi-user collaboration
+ Shared GPU training
+ Framework-neutral adapters
+ Local-first deployment

Computer vision, text, tabular, audio, video, and LLM dataset workflows ship in this package. Visual pipelines, Team/Enterprise setup, and org→workspace→project hierarchy are included. See the docs roadmap for planned SSO handshake and shared GPU scheduling.

Install

pip install oneopen-ml-studio

With extras:

pip install "oneopen-ml-studio[yolo]"       # Ultralytics Label Assist + training
pip install "oneopen-ml-studio[libreyolo]"  # LibreYOLO backend
pip install "oneopen-ml-studio[sam]"        # SAM deps (torch); install segment-anything separately
pip install "oneopen-ml-studio[all]"

From source:

git clone https://github.com/1-OpenSource/OneOpen-ML-Studio.git
cd OneOpen-ML-Studio
pip install -e ".[yolo]"

Quick start (local mode)

oneopen init
oneopen start          # alias: oneopen serve → http://127.0.0.1:8765

Or: python -m oneopen_ml_studio start

No cloud account. No mandatory registration. Data stays under ~/.oneopen (or ONEOPEN_DATA_DIR).

What works today

Capability	Status
Local CV: annotate, versions, YOLO / Torchvision train & export	Available
Image classification & segmentation catalogs	Available
Dataset profiling, quality checks, readiness score	Available
Experiment tracking + training adapters	Available
Text / tabular / audio / video / LLM workspaces	Available
Plugin manager (auto-install extras)	Available
Active learning selection	Available
Plugin registry + pipeline CRUD/runner	Available
Docker Compose (Postgres/Redis/MinIO)	Available
Multi-user orgs, review queues, Celery workers	Available (SSO handshake forthcoming)
Kubernetes enterprise starter	Available

See docs: Quality and Experiments tabs in the project UI, and /api/platform.

Platform modes (vision)

Mode	Users	Stack (target)
Local	Individuals, students, offline	FastAPI + UI, SQLite, local FS, optional GPU
Team server	Labs & startups	PostgreSQL, Redis, Celery, MinIO/S3, auth, workers
Enterprise	Regulated / multi-team	K8s, SSO (OIDC/SAML/LDAP), audit, HA, quotas

Documentation

https://oneopensource.org/oneopen-ml-studio/

Includes product vision, architecture, roadmap, package notes, and the full user guide.

pip install "oneopen-ml-studio[docs]"
cd docs && make html    # Windows: .\make.bat html

CLI

oneopen start                 Start local web UI (default :8765)
oneopen serve                 Same as start
oneopen init                  Initialize ~/.oneopen
oneopen version               Print version

License

Apache License 2.0 — see LICENSE and NOTICE.

@software{oneopensource_oneopen_ml_studio_2026,
  author = {oneopensource},
  title  = {OneOpen ML Studio},
  year   = {2026},
  url    = {https://github.com/1-OpenSource/OneOpen-ML-Studio},
  note   = {Open-source local-first ML data platform}
}

oneopensource — creator and maintainer.