Agents release notes

You can now use simulations to evaluate and refine your AI agents before they go live. Simulate both input data and agent tools directly during testing, at design time. This feature is available in the Test Configuration panel when you manually run your agent, and when you create evaluations and evaluation sets.

Use simulations to:

Simulations are ideal in early development for testing edge cases, when real tools are unavailable or costly to run, or if you want to avoid triggering changes in external systems.

For details, refer to the Evaluations section.

DeepRAG (Deep Research-Augmented Generation) is now available for Agents. This powerful new capability enables multi-document synthesis and retrieval for advanced agent use cases. With DeepRAG, agents can synthesize information from multiple documents, delivering comprehensive, citation-backed answers to complex queries. Note that this feature is currently supported only for PDF files.

DeepRAG queries are priced at 0.20 Platform Units (or 1 AI Unit) per request, supporting up to 20 pages or 30,000 tokens. Trial and Enterprise users receive 2 DeepRAG requests per day at no additional cost. Note that advanced ingestion is required for large, image-heavy, or tabular documents to ensure accurate synthesis. For licensing details, refer to Context Grounding licensing.

The UiPath LlamaIndex Agents SDK is now available. This Python SDK allows developers to seamlessly build and deploy LlamaIndex agents to the UiPath Cloud Platform. With this integration, developers can interact with UiPath Cloud services from within their LlamaIndex-based agents, and leverage Action Center integration to incorporate human-in-the-loop (HITL) workflows.

For details, refer to the Quickstart guide: UiPath LLamaIndex Agents.

We have updated the Agents page in Automation Cloud to provide clearer navigation and deeper visibility into agent activity.

The Instances section is now organized into three distinct tabs: All agents for a comprehensive view of all your agents, Early adopter agents for agents created as part of the early adopter program, and Errors (formerly Incidents) for quickly surfacing failed runs. The Management section includes Templates, for quickly accessing the Marketplace, and Indexes, which brings all the index monitoring capabilities from Orchestrator into a unified, accessible interface. For details, refer to Managing UiPath agents.

We have also introduced a new trace visualization feature accessible from the All agents instances tab. This view presents each agent run as an interactive node graph, allowing you to zoom, pan, and explore execution paths. Hovering or clicking on nodes reveals detailed information such as input/output data, logs, and performance metrics. For details, refer to Traces. Note that trace visualization is a preview feature.