EinsteinGptSettings in SFDX & Metadata API

Architectural Overview#

The EinsteinGptSettings metadata type represents the org-level configuration for Salesforce Generative AI (Einstein 1 Studio) (Setup > Einstein > Einstein Generative AI > Setup).

Generative AI in Salesforce empowers users with automated email drafting, case summarization, dynamic record insights, and conversational intelligence. Crucially, enterprise adoption of Large Language Models (LLMs) demands strict data privacy, enterprise trust, and regulatory compliance. Salesforce addresses these requirements through the Einstein Trust Layer, which provides dynamic data masking for Personally Identifiable Information (PII), strict zero-data-retention agreements with third-party foundation model providers (such as OpenAI, Anthropic, and Google Cloud), toxicity scoring, and audit trail logging.

EinsteinGptSettings serves as the global feature switch governing:

• Org-wide enablement of generative AI prompt execution and Prompt Builder templates (GenAiPromptTemplate metadata).
• Activation of Model Builder integrations connecting to custom foundation models via AWS Bedrock or Google Cloud Vertex AI.
• Audit trail logging and security telemetry across all generative LLM requests.

Directory and File Suffix Reference#

In Salesforce DX projects, EinsteinGptSettings is stored as a singleton file in the settings directory.

Key Metadata Fields#

The EinsteinGptSettings metadata type extends Metadata and exposes the following org-level properties:

Declarative XML, Scratch Org, and Package.xml Examples#

1. Settings Metadata File (EinsteinGpt.settings-meta.xml)#

<?xml version="1.0" encoding="UTF-8"?>
<EinsteinGptSettings xmlns="http://soap.sforce.com/2006/04/metadata">
    <enableEinsteinGpt>true</enableEinsteinGpt>
    <enableGenerativeAi>true</enableGenerativeAi>
    <enableTrustLayerAudit>true</enableTrustLayerAudit>
</EinsteinGptSettings>

2. Scratch Org Definition (config/project-scratch-def.json)#

To enable Generative AI in scratch org environments for automated testing of prompt templates and Apex ConnectApi calls:

{
  "orgName": "Generative AI CI/CD Testing Scratch Org",
  "edition": "Enterprise",
  "features": ["Einstein1PlatformUser", "SalesUser"],
  "settings": {
    "einsteinGptSettings": {
      "enableEinsteinGpt": true,
      "enableGenerativeAi": true,
      "enableTrustLayerAudit": true
    }
  }
}

3. Package Manifest (manifest/package.xml)#

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

Retrieval and Deployment Commands#

Manage EinsteinGptSettings via modern Salesforce CLI commands:

# Retrieve EinsteinGpt settings from target org
sf project retrieve start --metadata Settings:EinsteinGpt

# Deploy settings to staging environment
sf project deploy start --metadata Settings:EinsteinGpt --target-org staging-sandbox

# Validate settings deployment in production pipeline
sf project deploy validate --metadata Settings:EinsteinGpt --target-org prod-ci

Senior Architect Deployment Best Practices and Gotchas#

• Prerequisite for Prompt Template Deployment: Deploying GenAiPromptTemplate components (Prompt Builder templates) will fail with missing feature errors if enableEinsteinGpt is false in the target org. In release pipelines, sequence EinsteinGpt.settings-meta.xml before prompt template metadata.
• Data Cloud Infrastructure Dependency: When enableTrustLayerAudit is true, generation audit logs are stored in standard Data Cloud data lake objects. Ensure the target org has active Data Cloud provisioning, or audit logs will be buffered temporarily in platform queues.
• Token Consumption Monitoring: Generative AI features consume Einstein 1 Platform generative credit allowances. Enterprise architects must monitor consumption via the Digital Wallet in Salesforce Setup to prevent prompt generation failures when quotas are exhausted.
• Sandbox Refresh Strategy: Like other AI features, generative AI toggles may revert to disabled status upon sandbox refresh. Automate the re-enablement of generative AI by deploying EinsteinGpt.settings-meta.xml as part of your post-refresh automated configuration jobs.

Frequently asked questions

What does EinsteinGptSettings govern in Salesforce?

EinsteinGptSettings governs org-wide settings for generative AI features, including Prompt Builder, Model Builder, and Einstein Trust Layer data masking.

How is EinsteinGptSettings deployed via package.xml?

It is deployed under the Settings metadata type using the member name EinsteinGpt, referencing the EinsteinGpt.settings-meta.xml file in the settings directory.

What happens when enableTrustLayerAudit is enabled in EinsteinGptSettings?

It turns on comprehensive audit trail logging in Data Cloud, recording PII masking, toxicity scores, and prompt grounding data for compliance reviews.

Can EinsteinGptSettings be configured in scratch org definition files?

Yes. Adding einsteinGptSettings with enableEinsteinGpt set to true in project-scratch-def.json allows automated testing of generative AI prompt templates.