Agents release notes

Conversational agents are now GA, introducing a new way to interact with UiPath Agents through natural-language chat.

Built for dynamic, multi-turn conversations, conversational agents interpret context, maintain memory, and handle real-time interactions—making them ideal for helpdesk assistants, guided workflows, and natural-language interfaces that require contextual understanding and follow-up exchanges. In contrast, autonomous agents remain the best fit for single-turn, fully defined tasks such as document processing, data extraction, or automated report generation.

Both agent types share the same runtime and administrative capabilities: they can access tools to retrieve or update information across systems, analyze and ground content in context indexes, and are monitored centrally through Instance Management dashboards for performance, usage, and auditing.

For details, refer to Conversational agents.

You can now monitor the condition of your Context Grounding indexes using the new Index Health score, available in Agents Instance Management and in the Orchestrator index management tab. The score automatically evaluates ingestion reliability and query activity to flag degraded or unused indexes, helping you keep your data sources clean and up to date.

Indexes now follow the folder-only permissions model, where permissions are checked for the folder that the index is part of. Data source permissions are checked only at index creation.

Agents now support Gemini models, available in preview.

Enterprise users can use the UiPath-managed Vertex AI models, Gemini 2.5 Flash and Gemini 2.5 Pro, directly from the model dropdown in agent settings, or bring their own Vertex AI subscriptions through the LLM Gateway Bring Your Own (BYO) option. In Community organizations, the BYO setup is available directly in the Agents interface instead of the AI Trust Layer.

Gemini support comes with the following limitation: Integration Service activities are currently not available as agent tools when using Vertex AI models.

For details on data residency, refer to Data residency and supported models.

Agents now include built-in guardrails that automatically detect and block sensitive data exposure and prompt injection attacks. You can configure these guardrails directly from the agent settings to enhance security, compliance, and reliability with minimal setup.

This feature is currently available in preview and supported on specific licensing plans. For details, refer to Guardrails.