Workspace & Projects
A project is the unit of scope in Orinth. It owns its datasets, its training and testing runs, and the models promoted out of them — so image work and text work never bleed into each other.
Creating a project
From the Projects page, create a project and give it a name and a description. A first-run welcome walkthrough opens here the first time you arrive; see Guided Tours.
Once a project exists, every other surface in the app reads from it. Switching projects switches the datasets you see in Dataset Studio, the runs you see in Training, and the models available for Testing and Inference.
What a project holds
| Entity | Where it lives |
|---|---|
| Datasets | Dataset Studio, grouped by task |
| Dataset versions | Frozen, materialized snapshots under a dataset |
| Training runs | Training, with logs and artifacts per run |
| Testing runs | Testing, grouped into comparison runs |
| Promoted models | Model Library, alongside reference and uploaded models |
| Prediction history | Inference, with payloads and timings |
Provenance
Every surface makes its context visible: which project, which dataset, which split, which model, which run. A metric you are reading always names the thing it measured. This is deliberate — an evaluation number with no provenance is not evidence.
Project settings
Project settings cover the name, description, and deletion. Deleting a project removes its datasets, runs, and generated artifacts from local storage. This is not reversible, so the app asks you to confirm by name.
Global settings
Account-free by design — the Settings page carries machine-level configuration rather than a profile:
- Hugging Face token — required only for private or gated Hub models.
- Optional LLM assistance key — used by Data Recipes when you opt into LLM-assisted generation. Leave it unset and recipes run deterministically.
- Storage paths — where datasets, models, and runtime artifacts are written.