Testing your Agent

Learn how to test your Agent Component directly within its configuration, without running the full pipeline.

The Agent Component includes a built-in testing capability that allows you to validate your configuration without running the full pipeline. Using the testing area available next to the main configuration, you can quickly verify whether the component is set up correctly.

How to test the Agent Component

Run a simple test

To run a basic execution of the Agent Component:

1. Configure the component as needed.

2. Click Run.

After the execution completes, the Output Details section displays the following information.

Logs

All execution logs are grouped here, providing visibility into the component’s behavior. The logs include:

• Configuration: Provider and model settings used in the execution.

• Input: Input context details, including configured tools and content retrieval status.

• System: The System Message that guided the model’s behavior.

• User: The User Message exactly as received, in text, Markdown, or JSON.

• Tool Call: Details about any tool calls, including arguments and results.

• Input/Output Guardrail: Information about applied guardrails and their impact on the execution.

These logs make it easier to validate results and troubleshoot unexpected behavior.

Output

The execution result in JSON format. You can search for specific fields using JSONPath expressions.

Run one or more tests using variables

You can also run multiple tests simultaneously by simulating different scenarios with mocked variables. This is ideal for comparing how different prompts react to various data inputs, without waiting for data from previous connectors.

To mock variables, your System or User prompts must include Double Braces expressions, such as:

{{ message.body.text }}

In this example, {{ message.body.text }} can be mocked with the following structure:

{
  "body": {
    "text": "This product is amazing! Best purchase ever."
  }
}

To configure mocked variables, follow these steps:

1. In the Playground of the Agent Component, click Experiments.

2. Click Select a dataset and choose Create a new dataset.

3. Enter a name for the dataset. You can create multiple datasets to test different scenarios.

4. Click New Experiment. You can create multiple rows (experiments) to test different variable values at once.

5. Select the desired expressions and provide their values in the Value field.

6. Select the Experiments you want to run and click Run.

The Agent Component automatically detects all Double Braces expressions used in the prompts.

How to view and save test results

All executions (Simple tests and Experiments) are automatically captured in the Traces section. Saving executions allows you to maintain a complete history for future reference and performance auditing. See the data retention and storage rules below:

• Auto-save: By default, executions are temporarily stored for 5 days. If you close and reopen the component, they will remain available within this window.

• Persistent storage: To prevent an execution from being deleted after 5 days, you must open the Traces section, select the desired executions and click Save.

In Traces, each execution is listed with the following details:

• Execution ID

• Execution status

• User Message used

• Latency (in seconds)

• Date and time

To save a trace and maintain the execution history, the pipeline must be saved beforehand.