December 2025

December 19, 2025

DOM extraction support for desktop applications

We've added DOM extraction support for desktop applications. This may cause a difference in behavior as compared to previous versions, when only image was used for desktop applications. In the ScreenPlay activity, you can use the Use DOM when available property to disable DOM extraction and revert to using image only.

For more information, refer to ScreenPlay configuration.

December 12, 2025

Trace files now display variable images

ScreenPlay trace files now render image variables directly in the HTML output instead of displaying their Base64-encoded data. This improvement makes trace files easier to read and visually inspect when reviewing execution logs or debugging ScreenPlay runs.

Official model rename in LLM Gateway

ScreenPlay now lists models in the Model drop-down using their official names, making it clear that the ScreenPlay activity runs on the UiPath Screen Agent, UiPath’s proprietary agentic framework for reasoning about and interacting with user interfaces.

The UiPath Screen Agent orchestrates actions and decisions across applications using large language models (LLMs) hosted through the LLM Gateway.

To learn more, refer to UiPath Screen Agent.

You can expect the following changes:

• Model names now follow the UiPath Screen Agent (powered by LLM name) format to show both the UiPath agentic layer and the underlying model provider.
• The previous format has been replaced by clearer and officially branded names.

The following table shows the old model name formats and their corresponding new model name formats currently displayed in the Model drop-down of the ScreenPlay activity:

Old model name	New model name
UiPath (with GPT-4.1)	UiPath Screen Agent (powered by GPT-4.1).
UiPath (with GPT-4 mini)	UiPath Screen Agent (powered by GPT-4.1 mini)
UiPath (with Gemini 2.5 Flash)	UiPath Screen Agent (powered by Gemini 2.5 Flash)
UiPath (with GPT-5)	UiPath Screen Agent (powered by GPT-5)
UiPath (with GPT-5 mini)	UiPath Screen Agent (powered by GPT-5 mini)
OpenAI – Operator	Azure OpenAI Operator
Anthropic – Computer Use	Anthropic Computer Use

Trace files output

The ScreenPlay activity now exposes a TraceFiles output object that returns the execution trace files generated at runtime. This enables automations to capture, store, or share execution trace files programmatically for debugging, testing, or audit purposes.

Bug fixes

We fixed an issue where in Studio 2024.10 and earlier, ScreenPlay did not correctly process inline images included in natural language prompts. Although the images appeared in the authoring interface, they were not expanded or transmitted to the ScreenPlay agent at runtime. As a result, any contextual information provided in those images was ignored during automation execution.