GenAiFunction Metadata Type Reference

The GenAiFunction metadata type represents an Agentforce Agent Action—the bridge between the Agentforce reasoning engine and executable Salesforce platform logic. Introduced in API version 60.0, a GenAiFunction wraps an underlying Flow, @InvocableMethod ApexClass, GenAiPromptTemplate, or Model Context Protocol (mcpTool) endpoint with natural-language instructions and JSON Schema definitions.

SFDX Bundle Directory Structure#

Unlike flat XML metadata types, each GenAiFunction is stored as a bundle directory under force-app/main/default/genAiFunctions/<Action_Name>/:

force-app/main/default/genAiFunctions/
└── Verify_Order_Return_Eligibility/
    ├── Verify_Order_Return_Eligibility.genAiFunction-meta.xml
    ├── input/
    │   └── schema.json
    └── output/
        └── schema.json

Key Metadata Fields#

Production-Grade Bundle Example#

1. Metadata XML (Verify_Order_Return_Eligibility.genAiFunction-meta.xml)#

<?xml version="1.0" encoding="UTF-8"?>
<GenAiFunction xmlns="http://soap.sforce.com/2006/04/metadata">
    <description>Checks whether a customer order is eligible for return based on the Order Number and purchase date. Always call this action before initiating a return.</description>
    <invocationTarget>OrderReturnEligibilityService</invocationTarget>
    <invocationTargetType>apex</invocationTargetType>
    <isConfirmationRequired>false</isConfirmationRequired>
    <isIncludeInProgressIndicator>true</isIncludeInProgressIndicator>
    <masterLabel>Verify Order Return Eligibility</masterLabel>
    <progressIndicatorMessage>Checking order return eligibility...</progressIndicatorMessage>
</GenAiFunction>

2. Input Schema (input/schema.json)#

{
  "unevaluatedProperties": false,
  "properties": {
    "orderNumber": {
      "title": "Order Number",
      "description": "The 8-digit alphanumeric Order Number provided by the customer (for example, ORD-94821).",
      "lightning:type": "lightning__textType",
      "lightning:isPII": false,
      "copilotAction:isUserInput": true
    }
  },
  "required": ["orderNumber"],
  "lightning:type": "lightning__objectType"
}

3. Output Schema (output/schema.json)#

{
  "unevaluatedProperties": false,
  "properties": {
    "isEligible": {
      "title": "Is Eligible",
      "description": "Indicates whether the order is within the 30-day return window.",
      "lightning:type": "lightning__booleanType",
      "copilotAction:isDisplayable": false,
      "copilotAction:isUsedByPlanner": true
    },
    "eligibilityReason": {
      "title": "Eligibility Reason",
      "description": "Explanation of why the order is eligible or ineligible for return.",
      "lightning:type": "lightning__textType",
      "copilotAction:isDisplayable": true,
      "copilotAction:isUsedByPlanner": true
    }
  },
  "lightning:type": "lightning__objectType"
}

Frequently asked questions

What is the SFDX directory structure for a GenAiFunction?

Each GenAiFunction lives in its own subfolder under genAiFunctions/ActionName/ containing ActionName.genAiFunction-meta.xml, an input/schema.json file defining expected inputs, and an output/schema.json file defining returned outputs.

Which invocationTargetType values are supported by GenAiFunction?

GenAiFunction supports apex, flow, generatePromptResponse (Prompt Builder templates), mcpTool (Model Context Protocol tools in API v65.0+), standardFunction,retriever, and datacloud.

What does requiresConfirmation do on a GenAiFunction?

When requiresConfirmation is true (API version 63.0+), the Agentforce UI prompts the human user to explicitly confirm the action before the underlying Flow or Apex method executes.

How does the LLM know what parameters to pass into a GenAiFunction?

The LLM inspects the GenAiFunction description and the property descriptions inside input/schema.json, where copilotAction:isUserInput marks parameters that the agent can extract from the conversation.

What is the purpose of progressIndicatorMessage on a GenAiFunction?

When isIncludeInProgressIndicator is true, the progressIndicatorMessage text (for example, Checking order eligibility...) is displayed in the chat window while the action is executing.