Building an agent in Studio Web

This section walks you through how to build an agent in Studio Web. Once you're familiar with the basics, check out Best practices.

Exploring the agent workspace

Use the main Studio Web designer canvas to design your agent, and the left and right panels to explore the agent structure and resources.

The left-side panel shows you the agent structure and includes:

• The Project Explorer – Organize and build your agent. Define prompts, tools, context, escalations, and more.

• The Data Manager – Define input and output arguments.

• The Errors panel – See design-time issues, broken configurations, or failed test runs.

The right-side panel includes:

• The Properties panel, split between:

  • The Properties tab – Configure agent-level and component-level settings.

  • The Dev tab – Run test inputs, debug, and inspect design-time traces.

• Agent score (Preview) – View your agent’s readiness based on evaluation results and test coverage.

• The Run output panel – Review results from the most recent test runs.

• Autopilot (Preview) – Get AI-powered suggestions to refine the prompts, tools, and agent setup.

Creating the agent

1. Go to studio.uipath.com.
2. Select the arrow next to the Create New button.
3. From the Create solution section, select Agent.
4. The new solution opens, with the Agent designer canvas displayed.

Configuring the agent

1. From the Project Explorer panel on the left, access the agent Definition, Evaluation sets, and Evaluators.
   Figure 1. The Agent inside a Solution
   

   

   
   
   1. The Definition panel is where you design and define the core elements of an agent. The definition contains the following sections: General, Tools, Contexts, and Escalations.
      • Use the General section to define the prompts for your agent.
      • Use the Tools section to connect runtime tools—like Integration Service connectors or published processes.
      • Use the Contexts section to link knowledge sources using Context Grounding indexes to give your agent relevant data access.
      • Use the Escalations section to set up human-in-the-loop fallbacks and enable Agent Memory to persist interactions and help calibrate future agent escalations.
   2. The Evaluation sets panel is where you create evaluations and store results. Evaluations objectively measure your agent's output and assess whether or not it is consistent with your objectives. For details, refer to Evaluations.
   3. The Evaluators panel is where you create and manage the evaluators used in evaluations.
2. First, select your agent under Project Explorer. Right-click to select Rename, and give your agent a unique name.
   The agent name helps to identify the agent across projects.
3. Next, access the Properties panel from the right-hand side menu to select the large language model you want to use with your agent.
   1. Select a model from the available dropdown list. Models are provisioned through UiPath AI Trust Layer.
   2. Configure the Temperature and Max. tokens per response fields:
   • Temperature determines the creativity factor of the LLM generation. You can set a value between 0 (Precise) and 2 (Creative).
   • Max. tokens per response refers to the maximum number of tokens to generate with the agent's response.
4. In the agent Definition panel, provide your agent with a System prompt and a User prompt.
   Add a well-structured prompt to give your agent instructions for what it should do, and constraints it should follow. For details, refer to Prompts.
5. Use the Data Manager panel to define the agent input and output arguments used in the prompts.
6. In the Tools section, select Add tool to add any tools your agent can use during execution.
   These can be a set of Integration Service activities, existing agents, and published processes to which you have access. Each tool must have a detailed description that helps inform the agent how and when to use this tool. For details, refer to Tools.
7. (Optional) In the Contexts section, select Add context to give the agent access to Context Grounding indexes.
   Choose an existing index available in your tenant, or select Create new to create a new index directly in Orchestrator. For details, refer to Contexts.
8. (Optional) In the Escalations section, select Add escalation to add a human in the loop through an escalation app and enable Agent Memory.
   Adding an escalation allows your agent to involve a human to review, approve, or update output as it is running. For details, refer to Escalations and Agent Memory.

Testing the agent

Now it's time to test your agent and see how you can improve it.

1. In the Properties panel, select the Dev tab.
2. Enter a sample input.
3. Select Test on cloud.
4. In the Test configuration window, confirm the resources used in the solution and the sample input.
5. Select Run.

You can then use the test output to refine the agent design.

Evaluating the agent

Next, go to the Evaluation sets and Evaluators panels to review and measure your agent.

1. In the Evaluation sets panel:

   • Rename the default evaluation set and add test cases with expected outcomes.

2. In the Evaluators panel, add evaluators to validate the agent output.

Each evaluation links a known input with an assertion about the output.

For details, refer to Evaluations.

Calculating the Agent score [Preview]

Select Open health score from the right-side panel to calculate the agent score. Refer to Agent score to learn how it is calculated.

Using Autopilot to improve your agent [Preview]

Select Open Autopilot from the right-side panel to receive suggestions on improving prompts, tools, and other components. Such improvements support a higher Agent Score and make your agent ready for deployment.

Publishing and deploying the agent

Once you have tested, evaluated, and refined your agent, it's time to publish it. This step is necessary for the agent to be available in the Run Job activity.

1. Select Publish to publish the agent to Orchestrator as a process. Select where to publish:
   1. Orchestrator Personal Workspace Feed - The process is published to your personal workspace and is only be available to you. This is the default selection.
   2. Orchestrator Tenant Processes Feed - The process is published to the global tenant feed and is available to all users in eligible folders.
2. Select a version. A version is generated automatically starting with 1.0.0 when you first publish, and then incremented automatically every time you republish.
3. Optionally, you can select Submit as a template to make the agent available as a template in your organization, or Share, to share it with users in your organization.

Running the agent

Using agents as activities means you can incorporate them into workflows to handle parts of larger deterministic processes, have multiple agents act in series, and start agent execution based on triggers.

To incorporate your agent into a workflow, follow these steps:

1. Publish your agent to Orchestrator.
2. In Studio, use the Run Job activity to integrate your agent into a workflow.

3. Whenever you make changes to your agent, republish the updated agent to Orchestrator and upgrade the corresponding process in Orchestrator.

   The Run Job activity always executes the version of the agent currently configured in Orchestrator. This ensures that your workflow consistently uses the most up-to-date version of your agent.

Refer to the System activities guide to learn how to configure and use the Run Job activity in your Studio workflows.

Note: Starting with UiPath.System.Activities version 25.4.2, the execution of agents and other Orchestrator jobs has been consolidated and improved under the Run Job activity. Existing workflows using the Run Agent activity continue to function without changes, ensuring a smooth transition. When adding new agent-running functionality to your workflows, the system automatically suggests and implements the Run Job activity, even if you search for Run Agent.