- Getting started
- Installation
- Installing ScreenPlay
- Best practices
ScreenPlay user guide
The ScreenPlay installation procedure is available in public preview, and supports the following deployment options:
-
The out-of-the-box (OOB) deployment option - gives you free access to all five available models within a free consumption package that you can spend on any of the models, as per your preference.
The quota renews on a monthly basis for the duration of the public preview.
The GA version of ScreenPlay will include other trial options.
-
The bring-your-own-model (BYOM) deployment option - allows you to use your own existing subscription to power ScreenPlay.
With BYOM deployment, there are no imposed limits on how much you can use the service.
To perform this installation, you need to have your own subscription for one of the following services:
- AWS Bedrock - Anthropic Computer Use
- Azure - OpenAI Operator
- Azure - OpenAI GPT 4.1
- Google - Gemini Flash 2.0
Note: The GA version of ScreenPlay will not require a separate subscription, all models will be available by default, covered by Agentic Units.
- Open UiPath Studio, and create a new project.
-
Add the latest UIAutomation package, 2025.10.5 or newer, by navigating to Manage Packages, then select Settings.
In the Default package sources section, ensure you enable the Official package.
- In the Manage Packages window, select All Packages, and then enable the Include prerelease filter.
- In the project canvas, select the Add Activity icon to search and add the ScreenPlay activity.
Note: For older versions, search and add the UI Agent activity.
-
In the ScreenPlay activity, add a prompt in the Task field.
Additionally, you can change the default LAM from the Model drop-down, and tailor other properties, if needed.
To configure ScreenPlay via Studio Web, take the following steps:
- Navigate to https://cloud.uipath.com, and log in to your account.
- Navigate to Automation Ops to create your Studio Web governance policy.
- In Automation Ops, in the Policies tab from the Governance view, select the governance policy you created.
-
Select the Activities tab, and then select the Enable preview and packages activities checkbox.
For more information, refer to Settings for Studio Web policies.
- Navigate to Studio, select Create New, and then select RPA Workflow.
- Select the Add activity icon to search and add the ScreenPlay activity to the project canvas.
-
In the ScreenPlay activity, add a prompt in the Task field.
Additionally, you can change the default LAM from the Model drop-down, and tailor other properties, if needed.
- In Automation Cloud™, navigate to Admin.
- Select AI Trust Layer, then select the LLM configurations tab.
- Select a tenant from the Tenant drop-down, and then select Add configuration.
- In the Add configuration panel, take the following steps:
- Select ScreenPlay from the Product drop-down.
- Select Replace UiPath LLM subscription from the Configuration mode drop-down.
- Select the subscription you have and want to use from the LLM drop-down.
- In the Configure customer managed connection section, select a folder from the Folder drop-down, and then select a connector from the Connector drop-down.
- Select Add new connection. A new tab opens, prompting you to provide information to connect UiPath to your selected connector. Populate the necessary fields, then select Test in the Add configuration panel.
- Select Save.
For a safe use of ScreenPlay, you must take the following important precautions:
- Current state-of-the-art LAMs are quite limited, therefore it is better to use multiple granular ScreenPlay activities, each handling one or two actions with a few target UI elements, rather than a single end-to-end automation instance attempting a complex task.
- Data collection — for the duration of the preview, we may collect data such as prompts, screenshots, the automated applications,
etc.
Important: Try ScreenPlay on non-sensitive information.
- ScreenPlay is a beta product, with a fair amount of agency and unpredictability.
Important: Use ScreenPlay on test environments only, with mock data. In case this scenario is not possible for you, ensure that the agent cannot do any real permanent damage in the sandboxed application where it runs.
