AIApplication Metadata Type in SFDX

Architectural Overview#

Across Salesforce’s predictive and generative AI stack—including Einstein Prediction Builder, Case Classification, Opportunity Scoring, and industry-specific AI—the platform organizes machine learning models inside an AIApplication container. Each AIApplication groups one or more MLPredictionDefinition and MLDataDefinition records under a unified lifecycle status.

Understanding and version-controlling AIApplication metadata allows architects to promote trained model definitions and writeback field mappings cleanly from staging sandboxes to production without rebuilding prediction schemas manually.

SFDX Directory & File Suffix#

Key Metadata Fields#

Declarative XML & package.xml Examples#

Sample .ai-meta.xml#

<?xml version="1.0" encoding="UTF-8"?>
<AIApplication xmlns="http://soap.sforce.com/2006/04/metadata">
    <developerName>Customer_Churn_Prediction_App</developerName>
    <masterLabel>Customer Churn Prediction App</masterLabel>
    <status>Enabled</status>
    <type>PredictionBuilder</type>
</AIApplication>

Sample package.xml Manifest#

<?xml version="1.0" encoding="UTF-8"?>
<Package xmlns="http://soap.sforce.com/2006/04/metadata">
    <types>
        <members>Customer_Churn_Prediction_App</members>
        <name>AIApplication</name>
    </types>
    <version>68.0</version>
</Package>

SOQL Query & SFDX CLI Commands#

Query Installed AI Applications via SOQL#

SELECT Id, DeveloperName, MasterLabel, Type, Status, CreatedDate, LastModifiedDate
FROM AIApplication
ORDER BY CreatedDate DESC

Retrieve and Deploy via Salesforce CLI#

# Retrieve all AIApplication definitions
sf project retrieve start --metadata "AIApplication:*"

# Deploy a specific AIApplication to the target org
sf project deploy start --metadata "AIApplication:Customer_Churn_Prediction_App"

Senior Architect Best Practices & Gotchas#

• Deploy Writeback Custom Fields First: If your AIApplication and child MLPredictionDefinition write scores back to a custom AI Prediction field (CustomField with <isAiPredictionField>true</isAiPredictionField>), include that CustomField in the same deployment package.
• Minimum Data Volume for Training: Deploying an AIApplication with <status>Enabled</status> to a partial sandbox or scratch org with fewer than 400 historical positive/negative examples will cause model training to fail post-deployment. Deploy as Draft in low-data environments.

Frequently asked questions

What is the AIApplication metadata type in Salesforce?

AIApplication is the foundational metadata container for Einstein predictive AI applications, such as Einstein Prediction Builder, Discovery integrations, and industry AI models.

Where are AIApplication files stored in SFDX?

They are stored in the force-app/main/default/aiApplications directory with the .ai-meta.xml or .aiApplication-meta.xml suffix depending on the CLI decomposition.

How does AIApplication relate to AIApplicationConfig?

AIApplication defines the core AI application identity and type, while AIApplicationConfig stores the operational runtime parameters and feature settings for that application.

Can I query AIApplication records via Tooling API or SOQL?

Yes, you can query the AIApplication object in SOQL to inspect DeveloperName, MasterLabel, Status, and Type across your org.