Agents release notes

DeepRAG (Deep Research-Augmented Generation) is now generally available for Agents. This 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. This feature is currently supported only for PDF files.

The DeepRAG option is available as a new search strategy when configuring contexts for an agent. Results returned by DeepRAG include source citations, which are automatically displayed in the agent trace for full transparency.

For details, refer to Working with DeepRAG.

We've redesigned the Agent Instance Management page to offer a more streamlined, unified experience for managing agents across their entire lifecycle.

What’s new:

This update delivers clearer organization, deeper analytics, and a more consistent management experience for every stage of agents lifecycle, from draft to deployment.

For details, refer to Managing UiPath Agents.

You can now add IXP models as tools in your agents. Agents can classify and extract data from structured, semi-structured, or unstructured documents at runtime using any published IXP model.

With IXP as a tool, agents can automatically interpret document content, extract high-confidence data points, and use these values in reasoning or follow-up tool calls. You can also connect Document Validation Station as an optional escalation step, allowing human reviewers to validate or correct extracted data before or after execution.

This feature is now available in public preview. For details, refer to IXP.

The Automation Ops governance policy for Studio Web includes new agent design policies to enforce consistency and quality in agent configurations. The new options, currently available in preview, allow you to enforce key settings before agents are published, including minimum agent scores, maximum temperature ranges, and token limits per response. By setting these parameters, you ensure that agents meet predefined standards and reduce the risk of errors or performance issues.

The governance options give you more control, ensuring that new agents comply with the required thresholds. For existing agents, any updates will prompt necessary adjustments to meet the policy settings.

For details, refer to Settings for Studio Web policies.

Agents now support Customer-Managed Key (CMK) encryption, allowing organizations to encrypt sensitive LLM telemetry data, including input and output data within traces, using their own encryption keys. This ensures full control over data protection, lifecycle management, and access policies, aligning with enterprise-grade security and compliance standards.

You can manage encryption and security policies directly in Admin > Settings > Security, where you can choose between UiPath-managed keys or Customer-Managed Keys. Enablement is done per organization and requires a support ticket request.

For details, refer to Encryption.

The Context Grounding tool in Agents now supports a Query/Prompt field for all search strategies. You can provide a static value or reference an agent argument that will be resolved at runtime. This allows you to deterministically control what the search strategy uses, rather than relying on the agent to infer it automatically. The Query field is available for Structured and Semantic search strategies, while the Prompt field applies to the DeepRAG strategy.

When using contexts for agents, you can scope semantic and structured searches to specific folders. This improvement enables more precise context retrieval by limiting searches to folder paths. Folder scoping is ideal for use cases where data is organized into subfolders, allowing agents to search only within the relevant context.

You can now add activities as tools by browsing nearly the entire Integration Service activities library. This enhancement gives you access to a broad range of API-based activities, expanding the variety of tools available to your agents.

All API-based Integration Service activities are supported. Triggers and other non-API activity types are not supported.

Activities that require mandatory file inputs or outputs, such as file download actions, are currently excluded. However, activities with optional file fields are supported, with the limitation that those specific file-related fields are automatically omitted from the tool configuration.

For details, refer to Activities.