The AI agent (create/edit) panel features two main sections: the left side for configuring the agent and the right side for testing and previewing its behavior. Use the Preview pane to test your AI agent before using it in a flow. You can run the agent with mock data, chat with it using your own inputs, compare responses across models, and review execution details to refine your instructions.
To see how import requests will look during configuration, select the Input tab.
Note
With Monitor access to the flow, you can use Preview to view the agent's responses. However, Chat mode is not available. For monitor privileges preview is available but Chat mode is not available. For Manage role chat mode is available.The Input tab displays the output received from the previous flow step. To provide your own sample data, use the Mock response tab. For more information, see Provide mock response data for imports.
Select Agent output at the right to test your agent and chat with it directly. The system uses the same configuration that will run in the flow, so you can safely validate behavior before running it in production.
Important: If the prompt used needs a websearch or image generation, you must select the required tools from Add tools section.
Select Test to run the agent using the configured mock input.
After the test completes, the response appears in the conversation pane for you to review the output and decide whether to refine the agent's instructions, model, or settings. Post response, you can do the following:
-
View used AI credits: If your agent uses Celigo AI, the preview also displays the estimated AI credits consumed for the most recent test. If your agent uses Bring Your Own Key (BYOK), the test runs through your configured model connection instead.
Note: The displayed AI credits apply only to the most recent test or conversation and are not a running total for the conversation.
-
View conversational history: You can view your chat and test history records for the selected time period. Click Clear history to eliminate the history details.
- Start a new conversation: Click the + icon or the Test button to clear the current conversation and start a new one.
- Regenerate response: This option generates a new response for the same input. Use this option if you're not satisfied with the current response or if the previous response was incomplete.
- Copy response: This option copies the response as plain text. If the response contains JSON or other structured content, you can paste and reuse it in your workflow or application as needed.
You can select Stop at any time to cancel a running test.
Chat with your agent
In addition to running a test, you can chat directly with the agent by entering your own prompts in the chat box. This lets you try different scenarios without modifying the mock input.
You can also attach supported files, such as images, PDFs, or structured JSON documents, to provide additional context for the conversation. The send option becomes available only after you enter a message.
To attach a file:
- Click the Add file icon in the input section of the chat.
- In the Upload files, click Click here and select the files.
- Click Open.
Send button - Submit your message to the AI agent. While the agent is generating a response, the Send button changes to Stop button. Select Stop to cancel the current response.
For the send option to be available, you must enter a prompt in the input field. You can remove any attached file before sending it.
Both test results and chat responses appear in the same conversation. If you want to regenerate the response, click the regenerate response icon in the chat box. You can also click the + icon or Test button to start a new conversation, this action will end your existing chat session.
Review conversation history
Select the history icon to see previous conversations with this agent. In the history section, you can do the following:
- Open a saved conversation to see both the input it used and the agent's response
- Use Clear history to delete the conversations from here.
- For each conversation you can:
- Rename: Enter a new name and click Save
- Delete: This option permanently deletes a conversation
Important:
- Conversation history is available only after the agent has been saved. Conversations created before saving are temporary and are not retained
- Attached files are available only during the active conversation. If you reopen a saved conversation later, the files are not displayed, although the agent's responses remain available
- If you have Monitor (view-only) access to the flow, the chat, history, and new conversation controls are visible but unavailable
- The history details are retained only for 30 days. Data older than 30 days is automatically deleted
View response details
You can monitor the response progress in the conversation, as the agent generates a response.
Select Show details to view the execution trace for a response. The trace summarizes the tools the agent used and other execution details, making it easier to understand the response and troubleshoot unexpected results.
If the agent cannot access a required tool, for example, because the associated connection is unavailable, the response displays an error explaining the issue.
If the preview cannot start because the agent's input is unavailable, verify that the previous flow step completed successfully and that the required input data is available.
You can run comparisons using other models for evaluation of quality, latency, and cost. Agent output always uses the agent’s saved configuration for that aspect. Once you understand the details, make edits to fine-tune their instruction or try changing other settings.
In the Flow builder, for each agent, you can enable debugging to track requests and responses between integrator.io and the connected application. The logs can assist you when troubleshooting by enabling you to view errors that caused a particular record to fail. You can also access information that may help you determine what you need to fix in your agent to avoid errors or optimize your integration's performance. It is now available for all agent step types.
The debug functionality is similar to the debug mode in a flow.
Perform the following actions to start collecting debug logs for an agent:
-
Open an agent in a flow step you want to debug.
Note
To view the View debug logs option, you must first save and close an agent as this option is not available in Create an agent page.
- In the Edit an agent, click View debug logs in the upper right corner of the page.
-
Click Start debug to turn on the debugger. Debug logs will be collected for the period of time you choose from the drop-down list. Choose one of the following time increments for debug log collection and click Apply:
- Next 15 minutes
- Next 30 minutes
- Next 45 minutes
- Next 60 minutes
For more information, see Debug flow steps.