Clicking on Agents in The Lab opens the Agents Dashboard, where you can manage all the Agents created within a given Project.
Here you see a panel that displays each Agent as a card.
Each card shows the Agent’s name, a brief description, and its current status. The status indicates whether the Agent is available in the Workspace for consumers:
- Draft: the Agent is still being configured and has not been published yet. It is displayed with a dotted circle icon.
- Unpublished: the Agent was published before, but changes were made and those updates have not been published yet. It is displayed with a dotted circle containing a check mark.
- Published: the Agent is active and available for consumers in the Workspace. It is displayed with a solid check mark inside a circle.
Any changes you make to an Agent are automatically saved as an Unpublished version. This ensures your progress is preserved even if you haven't published the Agent manually.
Clicking on an existing Agent opens a detailed panel on the right side of the screen. This panel displays additional information about the selected Agent, including its full description and AI model, among other details.
Each Agent displayed includes both a Test button and an Edit button.
Use Test to open the panel on the right and interact with the Agent, and use Edit to open a new tab where you can continue configuring that Agent. At the top of this new tab, you'll see the Agent's full name along with its description.
When you return to the Agents Dashboard, each Agent is displayed as a card featuring a More Options menu. This menu allows you to perform additional actions, such as:
-
Configures whether a consumer can share, via a link, the conversation they are having with this Agent in the Workspace (for example, None or Anonymous). For more details, see Sharing a conversation with an Agent with other consumers via a link.
-
Lets you define who can execute the Agent after it is published (for example, None or Project), according to the options described in Publishing an Agent to The Station.
-
Downloads a LabPackage containing:
- The Agent itself.
- Complete definitions of private resources (Tools or Agents) and public resources from the same Project that the Agent uses.
- References to external public resources used by the Agent.
This button is disabled until the Agent has been published at least once. If there are unpublished changes, the export will include only the last published version.
-
Permanently removes the Agent from the dashboard.
-
Publishes the Agent to The Station so consumers can discover it in the catalog and run it in the Workspace, according to the execution permissions you configured.
In the upper-right corner of the screen, you'll find the New Agent button. Clicking on this button offers three options:
-
Guides you step by step through creating a new Agent.
-
Allows you to manually configure all the details of a new Agent.
-
Enables you to import a LabPackage file containing an Agent along with all the resources it needs to function (such as Tools or other Agents included as resources). Selecting this option opens a panel where you can upload or drag and drop the file.
When importing, keep the following in mind:
- The process creates the Agent and all its resources. If the Agent or any of its resources already exist in the Project under the same name or ID, the import will fail. Importing does not overwrite existing resources; it only adds new ones.
- The package must be a LabPackage generated by the Export Agent feature.
- Any external public resources (other Agents or Tools) referenced in the package must already exist in the Project.
Regardless of whether you select "New Agent Assisted" or "New Agent Manually" to create your Agent, a new editing tab will always open by default.
On the left, you'll see the chat with Iris, the integrated Assistant, who you can interact with for help during Agent creation.
On the right, three main tabs are displayed: