> For the complete documentation index, see [llms.txt](https://docs.unitlab.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.unitlab.ai/documentation/get-started/the-unitlab-object-model.md).

# Key concepts

The building blocks of Unitlab (assets, datasets, projects, ontologies, workflows, tasks and releases) and how they fit together.

Every labeling job in Unitlab uses the same building blocks. Once you know them, every page of the app makes sense.

## How the pieces fit

```mermaid
flowchart LR
  A["Assets"] --> B["Dataset version"]
  B --> C["Project"]
  C --> D["Tasks"]
  D --> E["Release"]
```

1. You upload files. Each file becomes an **asset**.
2. You pick the assets you want to label and save them as a **dataset version**.
3. You attach that version to a **project**. The project has an **ontology** (what to label) and a **workflow** (who labels and reviews).
4. Each file becomes a **task** that moves through the workflow.
5. When the work is done, you export a **release**.

## Data

| Term           | What it is                                                                                                                                            |
| -------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Asset**      | One file stored in or connected to Unitlab: an image, video, PDF, scan and so on. You find assets under **Data Space** > **Assets**.                  |
| **Folder**     | A place to organize assets, like a folder on your computer. A folder can also mirror a path in your cloud storage.                                    |
| **Data Group** | Several related files that are labeled together as one task, for example front and side camera views of the same car.                                 |
| **Dataset**    | A named selection of assets or groups. You publish **versions** of it; a version never changes, so a project always knows exactly which files it got. |

## Labeling

| Term                     | What it is                                                                                                                                                                              |
| ------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Project**              | Where the labeling happens. It holds the data, ontology, workflow, instructions, queues and releases for one job. One project can mix data types.                                       |
| **Ontology**             | The list of things to label: **classes** (such as *Car*), the shape each class uses (box, polygon, mask...), and the **properties** annotators fill in (such as *color* or *occluded*). |
| **Item Property**        | A question about the whole file instead of one object, such as *weather: sunny / rainy*.                                                                                                |
| **Instructions**         | Your labeling rules, with examples, shown to annotators inside the project.                                                                                                             |
| **Annotation Workbench** | The editor where annotators draw and edit labels. It adapts to the data type.                                                                                                           |

## Teamwork and quality

| Term                      | What it is                                                                                                                                                       |
| ------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Workflow**              | The path each file follows, made of **stages** such as Annotate, Review and Complete. Stages can also be AI models, consensus votes or automatic quality checks. |
| **Task**                  | One file (or Data Group) in a workflow stage, assigned to a person or waiting to be picked up.                                                                   |
| **Queue**                 | The list of tasks waiting for you. The **Task Queue** shows work by stage; a **Batch Queue** tracks one upload.                                                  |
| **Comment** and **Issue** | Feedback pinned to a place on a file. An issue has an owner and is tracked until it is solved.                                                                   |

## Output

| Term              | What it is                                                                                            |
| ----------------- | ----------------------------------------------------------------------------------------------------- |
| **Release**       | A frozen, versioned export of your labels and files, with optional train, validation and test splits. |
| **Export format** | The file format of a release, such as COCO, YOLO, UUEF or GeoJSON.                                    |

## Workspace

| Term          | What it is                                                                                                                 |
| ------------- | -------------------------------------------------------------------------------------------------------------------------- |
| **Workspace** | Your team's space in Unitlab. It holds members, projects, data, models and settings. You can belong to several workspaces. |
| **Role**      | What a member can do, such as Owner, Manager, Annotator or Reviewer. Admins can create custom roles.                       |
| **AI model**  | A public or private model used to label data automatically, for example in a Model stage of a workflow.                    |

{% hint style="info" %}
**Tip:** When you share a result with your ML team, note the dataset version and release version. Together they tell exactly which files and which labels were used.
{% endhint %}

## Related

[Navigate the platform](https://docs.unitlab.ai/documentation/get-started/platform-navigation.md) · [Datasets overview](https://docs.unitlab.ai/documentation/datasets/datasets-overview.md) · [Workflows overview](https://docs.unitlab.ai/documentation/workflows/workflows-overview.md)
