Use of customer data with Agents

Data processing practices for UiPath Agents, covering the purposes for which customer data is collected and how it is used at runtime.

Purposes

We process the information used with our services for the following purposes:

• Provide our services
• Support our services
• Product improvement and development

UiPath processes customer data to deliver and improve its services, enhance product reliability, and provide a better user experience. We adhere to industry standards, legal requirements, and internal policies to ensure the responsible use and protection of customer data.

Provide our services

Agent creation and execution

During the creation and execution of agents, we process both design-time data (such as content within system and user prompts) and runtime data (such as traces with inputs, outputs, or tool calls). This enables agents to function as intended, in accordance with our Agent Builder documentation.

We do not store your data outside of data retention policies.

You can configure data guardrails (including Personally Identifiable Information (PII) masking, model restrictions, and bring-your-own-model) from the AI Trust Layer settings. Changes apply immediately and can be updated at any time.

Agent reliability

We do not use enterprise customers' data to train models.

To ensure agents perform reliably, and to help you build better agents over time, we store the following data during design time:

• Agent prompts and tools
• Debug runs
• Evaluation sets and evaluation runs

We process this information when you use features like input simulation, evals generation, Autopilot, prompt coverage, or prompt optimization. All features are opt-in.

Agent health score

We process the following data to calculate agent health metrics:

• Prompts
• Tools
• Input schema
• Evaluations

This helps identify and address potential issues, ensuring stable and predictable performance reflected in the agent health score. This is an opt-in feature.

Prompt optimization features

Prompts and evaluation runs are processed to improve prompt quality and agent effectiveness. This is an opt-in feature.

Account provisioning and telemetry

We process the following information to provision accounts and ensure secure, stable service delivery:

• Account data: Cloud organization name/ID, tenant name/ID, agent IDs, evaluation IDs, and user identity information.
• Usage data: Error logs and run traces to support performance improvements.

Support our services

Data submitted through support channels is processed to troubleshoot issues, in accordance with our Support Terms.

Data storage

Customer data is stored based on regional preferences and in accordance with our Data Residency page, available in the Automation Cloud admin guide.

Integration with generative AI services

• Prompts are processed by available LLM models to generate outputs. Agents require generative AI functionalities. You can bring your own model if you do not wish to use the ones available in the platform.
• Agents are designed to enable human escalation and oversight. You are responsible for making sure that human validation/human-in-the-loop is configured.
• Agent memory from human feedback is used to support agent output performance and is an opt-in feature.
• Third party LLMs managed by UiPath do not use customer data to train the models.
• You can govern and manage UiPath-managed generative AI features through the AI Trust Layer.

Safeguards

We follow our standard Software as a Service policies and related guidelines:

1. Customer data is deleted in accordance with the contractual provisions or upon request.
2. Customer data is encrypted. Access is strictly limited, and any access is logged and audited.
3. Data processed for product improvement is filtered and transformed to remove sensitive data or content.
4. You can export data from the AI Trust Layer Audit tab. For details, refer to Exporting audit logs.

We recommend you do not include sensitive data in the agent design time configuration.