How can I customize the output of an AI agent workflow tool?

Below is an overview answering your question regarding custom responses from workflows used as AI agent tools:

What Changed

In earlier versions you might have seen an option to directly set a custom output from an AI agent tool workflow. Recently, that direct option has been removed as part of recent improvements to how custom workflow tools work within n8n.

How to Work Around It

Instead of relying on a built‐in custom response field, you can now control the response by shaping the output within your workflow. Here are two approaches:

1. Using a Code (or Set) Node:
   Add a Code or Set node at the end of your sub-workflow that formats and returns your custom response exactly how you need it. This node can be used to create a JSON object with the output message. For example:

   JAVASCRIPTCopied
   // Code node example:
   response = { "response": "Your custom response goes here." };
   return response;

2. Building a Sub-Workflow for the Custom Response:
   If your workflow is acting as a tool for an AI agent (similar to the human fallback setup), you can create a dedicated sub-workflow that handles validating inputs (like checking for an email address) and formats the response accordingly. This sub-workflow is then called using the Custom n8n Workflow Tool node.

Summary

The removal of the direct custom response setting means you now manage the output within the workflow itself. You can either add nodes to format the response or build a sub-workflow to process and return exactly what you need. This method is fully supported and gives you more flexibility over how output is constructed.

Hope this helps clarify and provides a clear path forward for building your desired responses!