Semantic dataset

A semantic dataset is a named selection of project data prepared for RebelCore Agent. In the Tree interface, semantic datasets are also presented as trained models.

One project can have several semantic datasets for different purposes. For example, you might create one for payroll questions and another for workforce planning.

Before you train

Confirm that:

• The correct project is open.
• The required fields are selected in the Tree.
• The selection does not include unnecessary sensitive fields.
• The project has the intended LLM exposure level.
• You have Create Semantic Dataset permission.

Train a semantic dataset

1. Open a project and curate the Tree.
2. Select a node to open the blade.
3. Choose DATA.
4. Enter a meaningful name in Name.
5. Click Train Model.

You can continue navigating the Tree while training completes.

Status

Status	Meaning
TRAINING	RebelCore is preparing the selected data. It cannot be used in Agent yet.
READY	The semantic dataset can be selected in Agent.
FAILED	Training did not complete. Review the available error and try again after correcting the cause.

The DATA tab also shows when each item was last updated.

Open in Agent

When the status is READY:

1. Select Question in RebelCore Agent.
2. Confirm that the correct dataset is selected in Agent settings.
3. Set the response mode and privacy controls for the new session.
4. Ask a focused question.

You can also open Agent directly and choose the semantic dataset from the settings cog.

Use multiple semantic datasets

Create separate semantic datasets when:

• Different teams need different subject areas.
• A focused question needs fewer fields than a broad exploratory dataset.
• Data owners require distinct access or exposure choices.
• You want clear names for recurring Agent workflows.

In Agent, selecting more than one dataset applies the most restrictive exposure level among them.

Delete a semantic dataset

If your role has Delete Semantic Dataset permission, use the delete action on its card.

Caution

Deletion is permanent. Agent can no longer select the deleted semantic dataset. The source import, module, and project remain available.

Before deletion, confirm that no active Agent workflow relies on it and record any details required by your organisation’s change process.

Troubleshooting

• Train Model is disabled: select at least one field, enter a name, and confirm your permission.
• Training failed: review the source module and Tree selection, then retry with a valid selection.
• The item is missing in Agent: confirm that it is READY and that you still have access to its project.
• Agent gives irrelevant answers: check that the intended fields are selected and consider training a more focused semantic dataset.