Creating a project

A project is the container that ties everything together for one piece of work — typically a single engagement with a single customer. It’s the second stop in the RebelCore™ flow:

Dataset    →    Project    →    Tree    →    RebelCore™ Agent

You import datasets into a project, the project renders them as a Tree, and from the Tree you hand specs off to the Agent for inference. Each project has its own tree, its own import history, and its own access controls.

Before you start

You’ll need:

• A customer to attach the project to. Customers are managed by your administrator — if the right one doesn’t exist yet, ask them to create it.
• A name for the project. Pick something descriptive (e.g., Acme — Q4 2026 reconciliation).

Steps

1. Open the Projects page from the side menu.
2. Click New project in the top-right.
3. In the form, choose the customer from the dropdown and enter a project name. Fill in any remaining fields as needed.
4. Pick an access setting:
   • All Users — every user in the customer can open the project.
   • Restricted Access — only listed users can open the project. See Restricted access below.
5. Click Create.

You’ll be taken straight into the new project’s workspace.

Restricted access

When you pick Restricted Access, a dual-listbox appears so you can choose which users can open the project. The control has two columns:

• Available — every product user in your customer who isn’t already granted. Each pane has its own search box that filters by name or email.
• Granted — users who can open the project. Pick users in the left list, then press the → button (or double-click a row) to add them. Press ← (or double-click a row in the right list) to remove them.

The owner pin

The user creating the project is automatically pinned to the granted list with a ⭐ OWNER badge and cannot be removed. This means the creator always retains access to the projects they made — even if they later transfer day-to-day work to others.

A row is persisted in the project’s access table with is_owner = true so the rule is enforced at the database level, not just in the UI.

Customer Super Admin owners

If a Customer Super Admin creates a project, no owner row is persisted — Customer Super Admins always have implicit access to every project in their customer regardless of the access setting. The blade shows an informational “Customer Owner — implicit access” entry in the granted column to make this explicit.

When access setting is “All Users”

The owner pin is still recorded so we know who created the project. Every other user in the customer has access via the All-Users rule. If the project is later switched to Restricted Access, the owner stays pinned automatically.

What’s next

A new project starts empty. To populate it:

1. Create a dataset — upload your raw files into an import batch.
2. Create your dataset — build the silver-tier dataset module from your imported batch.
3. Open the Tree — your data now appears as labelled nodes you can navigate, refine, and act on.
4. Hand a node off to the Agent when you’re ready to run inference.

Modules — picking more than one

The MODULES section lets you tick one or more completed modules. The project’s tree merges across all selected modules — see How the Tree works for the merge behaviour.

Each module pill in the picker carries a small coloured badge showing its LLM exposure level — Full, Limited, or Advisory only. The level governs what the RebelCore™ Agent is allowed to say about that data when answering prompts. If your selected modules carry different levels, an inline notice appears under the picker explaining which level the project will adopt:

“Your selected datasets have different LLM exposure settings (Full × 1, Limited × 2). This project will run as Limited — the most restrictive setting wins.”

The project’s effective exposure is the most-restrictive level across its modules — Advisory beats Limited beats Full. There is no override; this is enforced at every prompt the agent runs against the project. See LLM exposure levels for the full rules and worked examples.

Modules are immutable after creation

Once the project is created, the module set is fixed. You can’t add or remove modules later. This is deliberate:

• Changing modules would invalidate any semantic dataset built from the project.
• It would silently change the labelled vector set the Agent queries against.
• It would break the audit trail for any prompt run against the project.

If you need a different module set, create a new project. The edit blade shows the modules in read-only form with a banner reminding you of the rule. The read-only view also surfaces the project’s effective exposure level so operators can verify the policy still matches what was intended.

Editing a project

Each card on the Projects page has an edit icon in its top-right corner (visible to anyone with the Add Projects permission). Clicking it opens the same blade you use for creating a project — but pre-populated with the project’s current values:

• Details — name, language, status, owner name, version, description (editable)
• Modules — current module selection (read-only — see above)
• Access — All Users / Restricted, plus the granted user list (editable)

You can change any editable field, add or remove granted users via the same dual-listbox, and Save Changes to persist. The owner pin still applies — even in edit mode the original creator stays in the granted list with the ⭐ OWNER badge and cannot be removed.

The card itself shows two small stat chips at the bottom-right: the number of users with access (1 if Restricted with just the owner, the full granted count otherwise) and the number of linked modules. Use these for a quick sense of the project’s footprint without opening it.

Next to the project name, the card also shows the project’s effective LLM exposure as a coloured pill — Full, Limited, or Advisory only. This lets you scan exposure posture across your portfolio at a glance. See LLM exposure levels for what each level means and how the effective level is computed.

Renaming or deleting a project

Renaming is done via the edit blade (above). Deletion is still an admin action — if you need to remove a project, contact your administrator.