UI Automation Activities

Last updated Dec 11, 2025

ScreenPlay

UiPath.Semantic.Activities.NUITask

Description

ScreenPlay is UiPath®’s next-generation automation agent, designed to bring agentic behavior and cognitive capabilities to the desktop. ScreenPlay interacts with applications much like a human would—navigating interfaces, adapting to change, and handling complex tasks that were previously infeasible with traditional automation methods.

To learn more, refer to ScreenPlay.

This activity must be added inside a Use Application/Browser activity.

Project compatibility

Windows | Cross-platform

Configuration

Task - Prompt describing the UI task to be performed, with the ability to:
- Use Variables
- Add image from screen (inline with the text)
- View last execution trace. For more details, see the Running and inspecting the execution results page.
Model - Indicates the underlying LLM used by ScreenPlay for task execution planning and reasoning. The following options are available:

UiPath (with Gemini 2.5 Flash)
- Basic model
- Works best on browsers
- Uses a proprietary implementation based on the page's DOM, using Gemini Flash for reasoning and image understanding
- Moderately fast
UiPath (with GPT-4.1)
- Standard model - for complex tasks
- Works best on browsers
- Uses a proprietary implementation based on the page's DOM and image understanding, using GPT-4.1 for reasoning
- Not very fast
UiPath (with GPT-4.1 mini)
- Basic model - faster, cheaper
- Works best on browsers
- Uses a proprietary implementation based on the page's DOM and image understanding, using GPT-4.1 mini for reasoning
- Moderately fast
UiPath (with GPT-5)
- Standard model - for complex tasks
- Works best on browsers
- Uses a proprietary implementation based on the page's DOM and image understanding, using GPT-5 for reasoning
- Slow
UiPath (with GPT-5 mini)
- Basic model - faster, cheaper
- Works best on browsers
- Uses a proprietary implementation based on the page's DOM and image understanding, using GPT-5 mini for reasoning
- Moderately fast
OpenAI - Operator
- Standard model - for complex tasks
- Works on any type of application, including image-based interfaces
- Uses OpenAI Operator, an image-based reasoning model. Likely the best of the bunch
- Slow
Anthropic - Computer Use
- Standard model - for complex tasks
- Works on any type of application, including image-based interfaces
- Uses Anthropic Computer Use, an image-based reasoning model
- Slow
Rate this activity - Good or Poor

Additional options

Options

Max number of steps - This is the maximum number of steps that ScreenPlay can take to achieve its goal. The property can be used as a basic guardrail to prevent infinite agentic loops.
Type by clipboard - Indicates whether the clipboard is used to type the given text. The following options are available:
- Never - Never use the clipboard
- Always - Always use the clipboard
- Whenever possible - Use the clipboard when possible. This depends on the OS and the text to be typed (e.g. If any special key is used, then the clipbaord will not be used)
Use DOM when available - Indicates whether DOM data will be used/sent to the LLM Model for applications where DOM can be extracted.

DOM can be used only by UiPath LAM implementations.

Disable if DOM-based targeting leads to incorrect element coordinates.

Default value is True.
Disable variable security - Indicates whether the variable security should be disabled.

Variable security ensures that prompt instructions can not be passed via variable values.

The purpose of this feature is to prevent prompt injection attacks and is based on LLM, so if a "false positive" occurs, the user can disable it for each ScreenPlay activity. To conclude if it is a "false positive", the user can inspect the execution trace, system prompt, reasoning, and actions.

Enable this option only if you need to pass prompt instructions via variable values or if a "false positive" result occurred.

Default value is False.
Input mode - Select which method should be used to generate keyboard and mouse input:
- Same as App/Browser - Uses the Input mode settings from the parent Use Application/Browser activity.
- Hardware Events - Acts as a real user by using "hardware" inputs such as mouse movements or keyboard strokes to interact with applications. These are hardware-triggered events sent directly to the operating system. While this method offers 100% behavioral emulation, some events may occasionally be lost. As a developer, it's your responsibility to ensure that all events reliably reach the target application.
- Chromium API - Performs actions using debugger APIs. Works only for Chromium elements. Sends all text in one go. Works even if target app is not in focus. For more details, refer to Chromium API.
- Simulate - Simulates using accessibility APIs. Recommended for browsers, Java based applications, SAP. Usually more reliable than Hardware Events. Sends all text in a single action. Works even if target app is not in focus. Please test if your target application UI element supports this.
- Window messages - Simulates using Win32 messages. Recommended for desktop apps. Usually more reliable than Hardware Events. Sends all text in one go. Works even if target app is not in focus. Please test if your target application UI element supports this.

Continue on error - Specifies if the automation should continue even when the activity throws an error. This field only supports Boolean values (True, False). The default value is False. As a result, if the field is blank and an error is thrown, the execution of the project stops. If the value is set to True, the execution of the project continues regardless of any error.

Timings

Delay before - Delay (in seconds) between the time the previous activity is completed and the time this activity begins performing any operations. The default value is 0.2 seconds. Adding a delay between activities ensures that one activity has enough time to complete before the next activity begins.
Delay after - Delay (in seconds) between the time this activity is completed and the time the next activity begins any operations. The default value is 0.3 seconds. Adding a delay between activities ensures that one activity has enough time to complete before the next activity begins.
Timeout - Specify the amount of time (in seconds) to wait for the activity to be executed before throwing an error. The default value is 30 seconds.

Output