Test a specialized agent

Test your specialized agent in Optimizely Opal before deploying it or after editing it. Testing lets you verify that your agent responds correctly and catches configuration issues before users encounter them.

Tests run inside the agent editor. Single-shot agents run once and return a result. Multi-turn agents let you give follow-up instructions after the agent responds.

Access your agents

1. Go to home.optimizely.com.
2. Select your organization.

3. Click Opal.

   

4. Click Agents.

   

5. Click the Your Agents tab.

   

Test your specialized agent

Follow these steps to run a Test Run on your specialized agent:

1. Follow the steps in the Access your agents section.
2. Click the agent name or click More actions (...) > Edit Agent.

3. Click Test Run.

   

4. (Optional) Enter values for your agent's input variables (placeholders defined in the agent prompt that are filled in at runtime). To remove all values, click Clear All.
5. Click Run.

The agent runs and displays the Execution Results. Click Stop Execution to end the Test Run. 

Test a multi-turn specialized agent

When you test a multi-turn specialized agent, the test run stays active after each response. This lets you continue the conversation across multiple turns.

1. Follow the steps in the Access your agents section.
2. Click the agent name or click More actions (...) > Edit Agent.
3. Click Test Run.

4. (Optional) Enter values for any input variables, then click Run.

   

5. Wait for the agent to complete Turn 1. An Agent response complete notification displays when the agent is ready for your reply.

   

6. Type your reply in the Type your message field and click Send to continue the conversation.

   Input variable values lock when the conversation starts. Click Reset to edit to change them and restart the test run.
7. Continue replying across as many turns as needed. The Output section updates to show Execution Results (Turn X) after each turn. Execution Memory grows to reflect the full conversation history.
8. Click Done to end the test run.

Execution results

The Execution Results include the following information:

• Output – The agent's execution results for the current turn.
  • Execution ID – The unique identifier of the request.
  • Status – Whether the run was successful.
  • Agent's response – The agent's text response for the current turn. For single-shot agents, this is the final result returned to Opal Chat or a workflow agent. For multi-turn agents, the label displays as Execution Results (Turn X) and updates with each turn.
• Execution Memory – Memory shows each step that Opal takes while executing the agent. Use this information to debug your agent and confirm that the outputs and steps are correct.
  • The first message shows the initial request, including the prompt template and variable values.
  • The last message is the final output returned to you.
  • Other messages in between are steps Opal takes to form its response, including tool calls in JSON format.

  • For multi-turn agents, memory grows with each turn and includes both User and Assistant chunks for the full conversation history.
     

    

The execution details also display in the specialized agent's logs.