Data Model for Generative AI Audit and Feedback

Understand and analyze your Einstein generative AI audit and feedback data by using the appropriate data model objects (DMOs). Einstein generative AI Audit and Feedback data is streamed to a data lake object (DLO) in Data 360 and mapped to data model objects (DMOs).

Required Editions

Available in: Enterprise, Performance, and Unlimited Editions with an Einstein for Sales, Einstein for Platform, Einstein for Service, Einstein 1 Service, or Einstein GPT Service add-on. To purchase add-ons, contact your Salesforce account executive.

Generative AI Audit and Feedback FlowChart

An entity relationship diagram showing the data model for generative AI audit and feedback.

Custom and Standard DMOs

Data 360 Data Lake Objects (DLOs) that contain generative AI audit and feedback data are mapped to custom DMOs (legacy) and standard DMOs.

Note Build dashboards, reports, and queries using the standard DMOs. Salesforce will implement future updates to Audit and Feedback schema in standard DMOs but not custom DMOs.

DLO	Custom DMO (Legacy)	Standard DMO
`GenAIAppGeneration`	`GenAIAppGeneration`	`Ai Response App Generation`
`GenAIContentCategory`	`GenAIContentCategory`	`Ai Content Quality Category`
`GenAIContentQuality`	`GenAIContentQuality`	`Ai Content Quality`
`GenAIFeedback`	`GenAIFeedback`	`Ai Feedback`
`GenAIFeedbackDetail`	`GenAIFeedbackDetail`	`Ai Feedback Additional Info`
`GenAIGatewayRequest`	`GenAIGatewayRequest`	`Ai Gateway Request`
`GenAIGatewayRequestTag`	`GenAIGatewayRequestTag`	`Ai Gateway Request Tag`
`GenAIGatewayResponse`	`GenAIGatewayResponse`	`Ai Gateway Response`
`GenAIGeneration`	`GenAIGeneration`	`Ai Response Generation`
`GenAIGtwyRequestMetadata`	`GenAIGtwyRequestMetadata`	`Ai Gateway Req Additional Info`
`GenAIGtwyRequestLLM`	`GenAIGtwyRequestLLM`	`Ai Gateway Request Model Diagnostic`
`GenAIGtwyObjRecord`	`GenAIGtwyObjRecord`	`Ai Gateway Request Object Record`
`GenAIGtwyObjRecCitation`	`GenAIGtwyObjRecCitation`	`Ai Gateway Req Object Record Citation`

Einstein Generative AI App Generation DMO

A Data 360 DMO that captures feature-specific changes made to the original generated text. The corresponding standard DMO is GenAiResponseAppGeneration (developer name: GenAiResponseAppGenerationDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Feature`	`feature__c`	Text	The AI feature associated with the gateway request. Example values: Service Replies, Sales Summaries
`Generation Id`	`generationId__c`	Text	Foreign key to the `GenAIGeneration` object.
`Generation Update`	`generationUpdate__c`	Text	Response text generated by apps after receiving a generation.
`Generation Update Id`	`generationUpdateId__c`	Text	Unique key submitted by app teams to identify app-specific mutations of a generation.
`Id`	`id__c`	Text	Primary key: The unique ID to the `GenAIAppGeneration` object.
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.

Einstein Generative AI Content Category DMO

A Data 360 DMO that captures Einstein Trust Layer detector result values by detector type and sub-category. This DMO includes the safety and toxicity scores of output from the LLM. The corresponding standard DMO is GenAiContentQualityCategory (developer name: GenAiContentQualityCategoryDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Category`	`category__c`	Text	Detection category. Example values for `PII` category: Name, Email, Credit Card Number. Example values for `Toxicity` category: Hate, Violence Example values for the Prompt Defense category: 0, 1 Example values for `InstructionAdherence`: `HIGH`, `LOW`, `UNCERTAIN` Example values for `TaskResolution` (Beta) Evaluation Result: `FULLY_RESOLVED`, `PARTIALLY_RESOLVED`, `UNRESOLVED`
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Detector Type`	`detectorType__c`	Text	The type of detector. Example values: `PII` `TOXICITY` `PROMPT_DEFENSE` `InstructionAdherence` `TaskResolution` (Beta)
`Id`	`id__c`	Text	Primary key: The unique ID to the `GenAIContentCategory` object.
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Parent`	`parent__c`	Text	Foreign key to the `GenAIContentQuality` or `GenAIGeneration` object.
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.
`Value`	`value__c`	Text	Detector value. The value depends on the detector type. Example values: When the detector type is PII and the category is email, the value indicates the number of email addresses detected in the prompt. When the detector type is toxicity, and the category is hate, the value indicates the likelihood of hate content in the response. When the detector type is `TaskResolution` (Beta), the value is the explanation text. Example: `User intent was to...`

Einstein Generative AI Content Quality DMO

A Data 360 DMO that captures whether a request or response is safe or unsafe. The corresponding standard DMO is GenAiContentQuality (developer name: GenAiContentQualityDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Content Type`	`contentType__c`	Text	Identifies whether the content is input (prompt) to the LLM or output (response) from the LLM. Values: INPUT or OUTPUT
`Feature`	`feature__c`	Text	The AI feature associated with the gateway request. Example values: Service Replies, Sales Summaries
`Id`	`id__c`	Text	Primary key: The unique ID to the `GenAIContentQuality` object.
`Is Toxicity Detected`	`isToxicityDetected__c`	Text	Detection of toxicity in the response. Values: True or False
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Parent`	`parent__c`	Text	Foreign key to the `GenAIGatewayRequest` or `GenAIAppGeneration` object.
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.

Einstein Generative AI Feedback DMO

A Data 360 DMO that captures feedback for a specific generation or part of a generation. The corresponding standard DMO is GenAiFeedback (developer name: GenAiFeedbackDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Action`	`action__c`	Text	The action taken by the user. Example values: generation-edit, reject, accept, thumbs-up, thumbs-down, regeneration.
`App Type`	`appType__c`	Text	The application associated with the gateway request. Also known as the `TargetId`. Example values: ReplyRecommendations
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Feature`	`feature__c`	Text	The AI feature associated with the gateway request. Example values: Service Replies, Sales Summaries
`Feedback`	`feedback__c`	Text	Feedback Values: `GOOD`, `BAD`, or `NULL` The value for the feedback field can be `NULL` when the user provides a full verbatim feedback.
`Feedback Id`	`feedbackId__c`	Text	Primary key: The unique ID of the `GenAIFeedback` object.
`Generation Group Id`	`generationGroupId__c`	Text	The generation group ID.
`Generation Id`	`generationId__c`	Text	Foreign key to the `GenAIGeneration` object.
`Generation Update Id`	`generationUpdateId__c`	Datetime	Foreign key to the `GenAIAppGeneration` object.
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Source`	`source__c`	Text	Source of the feedback. Values: `HUMAN`, `SYSTEM`, or `NOT_PROVIDED`
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.
`User Id`	`userId__c`	Text	Salesforce user ID.

Einstein Generative AI Feedback Detail DMO

A Data 360 DMO for details of user feedback. This DMO can include sensitive user data. The corresponding standard DMO is GenAiFeedbackAdditionalInfo (developer name: GenAiFeedbackAdditionalInfoDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`App Feedback`	`appFeedback__c`	Text	Mapping of key-value pairs for context on the action taken by the user. The option and user selection are represented as key-value pairs. Example values: Inaccurate-True, Incomplete-True
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Feature`	`feature__c`	Text	The AI feature associated with the gateway request. Example values: Service Replies, Sales Summaries
`Feedback Detail Id`	`feedbackDetailId__c`	Text	Primary key: The unique ID to the `FeedbackDetail` object.
`Feedback Text`	`feedbackText__c`	Text	The full verbatim feedback text provided by the user. Example value: `This email didn't include all the relevant information.`
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Parent`	`parent__c`	Text	Foreign key to the `GenAIFeedback` object.
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.

Einstein Generative AI Gateway Request DMO

A Data 360 DMO that captures prompt input, request parameters, and model details and parameters. This DMO can include sensitive data. The corresponding standard DMO is GenAiGatewayRequest (developer name: GenAiGatewayRequestDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`App Type`	`appType__c`	Text	The application associated with the gateway request. Also known as the `TargetId`. Example values: Service Replies, Sales Summaries
`Bot Version Id`	`botVersionId__c`	Text	Bots version ID.
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Completion Tokens`	`completionTokens__c`	Number	The number of tokens used to complete the response generation by the LLM. The LLM provider returns this metric. Example values: 21, 13
`Enable Input Safety Scoring`	`enableInputSafetyScoring__c`	Text	Flag to turn safety scoring for input on or off. Values: TRUE or FALSE
`Enable Output Safety Scoring`	`enableOutputSafetyScoring__c`	Text	Flag to turn safety scoring for output on or off. Values: TRUE or FALSE
`Enable Pii Masking`	`enablePiiMasking__c`	Text	Specifies whether to mask certain types of sensitive data such as personally identifiable information (PII). If set to true, data masking is turned on.
`Feature`	`feature__c`	Text	The AI feature associated with the gateway request. Example values: Service Replies, Sales Summaries
`Frequency Penalty`	`frequencyPenalty__c`	Number	The penalty that's applied to the repetitiveness of generated tokens. The penalty applied uses the value provided and is proportional to the number of times the tokens have already appeared in the prompt or in prior generations. The minimum value is `0.0`. The maximum value is `1.0`.
`Gateway Request Id`	`gatewayRequestId__c`	Text	Primary key: The unique ID to the `GenAIGatewayRequest` object.
`Generation Group Id`	`generationGroupId__c`	Text	The generation group ID.
`Masked Prompt`	`maskedPrompt__c`	Text	Masked prompt or input text. Example: Generate an email to PERSON_0 inviting them to an event near them.
`Model`	`model__c`	Text	The name of the model associated with the request. Example values: gpt-35-turbo, gpt-4
`Num Generations`	`numGenerations__c`	Number	Number of expected generations. Example values: 1, 2
`Org Id`	`orgId__c`	Text	Salesforce ID.
`Parameters`	`parameters__c`	Text	Mapping of key-value pairs of the request parameters. Request parameters specify the user-configured settings for the LLM to use when generating a response.
`Planner Id`	`plannerId__c`	Text	Bots planner ID.
`Presence Penalty`	`presencePenalty__c`	Number	Similar to `Frequency Penalty`, used to reduce the repetitiveness of generated tokens. However, the presence penalty is applied equally to all tokens that have already appeared, regardless of how many times they’ve appeared. The minimum value is `0.0`. The maximum value is `1.0`.
`Prompt`	`prompt__c`	Text	The resolved version of the prompt text before data masking is applied to the text. If data masking is turned on, the actual prompt sent to the LLM has sensitive data masked.
`Prompt Template Dev Name`	`promptTemplateDevName__c`	Text	The ID of the prompt template.
`Prompt Template Version No`	`promptTemplateVersionNo__c`	Text	The version of the prompt template.
`Prompt Tokens`	`promptTokens__c`	Number	The number of tokens used in the prompt.
`Provider`	`provider__c`	Text	The name of the model provider. Example values: OpenAI, AzureOpenAI.
`Session Id`	`sessionId__c`	Text	Feature-specific session ID.
`Stop Sequences`	`stopSequences__c`	Text	An array or sequence of strings. The generated text is cut at the end of the earliest occurrence of a stop sequence. Stop sequences are useful to control the length and the structure of the response. Example: “\n” to stop the generation with a new line.
`Temperature`	`temperature__c`	Number	The temperature used for the generation. Higher values indicated that the model takes more risks. Lower temperatures mean that generations are less random. The minimum value is `0.0`. The maximum value is `1.0`.
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.
`Total Tokens`	`totalTokens__c`	Number	The total number of tokens used for the request. It’s the sum of the prompt and completion tokens. Example values: 25, 35
`User Id`	`userId__c`	Text	Salesforce User ID.

Einstein Generative AI Gateway Request Tag DMO

A Data 360 DMO that captures any custom data points as key-value pairs. The corresponding standard DMO is GenAiGatewayRequestTag (developer name: GenAiGatewayRequestTagDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Id`	`id__c`	Text	Primary key: The unique ID to the `GenAIRequestTag` object.
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Parent`	`parent__c`	Text	Foreign key to the `GenAIGatewayRequest` object.
`Tag`	`tag__c`	Text	The tag applied to the request for more context.
`Tag Value`	`tagValue__c`	Text	The request tag value. Example value: test
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.

Note If you query GenAIRequestTag data for analytics, auditing, or debugging, Salesforce enforces a maximum number of individual rows written per LLM request. Salesforce stores tags that exceed the limit into a single REMAINDER_TAGS row containing a JSON array. For more information, see “Tag Limits in the GenAIRequestTag DLO” in Share Insights from Einstein Generative AI Audit and Feedback Data.

Einstein Generative AI Gateway Response DMO

A Data 360 DMO that enables joins between the GenAIRequest DMO and GenAIGenerations DMO. The corresponding standard DMO is GenAiGatewayResponse (developer name: GenAiGatewayResponseDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Generation Request Id`	`generationRequestId__c`	Text	Foreign key to the `GenAIGatewayRequest` object.
`Generation Response Id`	`generationResponseId__c`	Text	Primary key: The unique ID to the `GenAIGatewayResponse` object.
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Parameters`	`parameters__c`	Text	Mapping of key-value pairs of the response parameters.
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.

Einstein Generative AI Generation DMO

A Data 360 DMO that captures generated responses returned for generation requests. This DMO also includes masked prompts if masking is turned on and can also include sensitive user data. The corresponding standard DMO is GenAiResponseGeneration (developer name: GenAiResponseGenerationDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Feature`	`feature__c`	Text	The AI feature associated with the gateway request. Example values: Service Replies, Sales Summaries
`Generation Id`	`generationId__c`	Text	Primary key: The unique ID to the `GenAIGeneration` object.
`Generation Response Id`	`generationResponseId__c`	Text	Foreign key to the `AIGatewayResponse` object.
`Masked Response Text`	`maskedResponseText__c`	Text	The LLM-generated response. If masking is turned on, this attribute can contain placeholder text. Example: `Hello Person_0, we would like to invite you to an event near you.`
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Response Parameters`	`responseParameters__c`	Text	Mapping of key-value pairs of the generation parameters. A response can have one or more generations.
`Response Text`	`responseText__c`	Text	The generated response after demasking. Example: `Hello John, we would like to invite you to an event near you.`
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.

Einstein Generative AI Gateway Object Record DMO

A Data 360 DMO that captures the information about the Salesforce objects that are related to Generative AI requests. The corresponding standard DMO is GenAiGatewayReqObjRec (developer name: GenAiGatewayReqObjRecDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Feature`	`feature__c`	Text	The AI feature associated with the gateway request. Example values: Service Replies, Sales Summaries
`Id`	`id__c`	Text	Primary key: The unique ID to the `GenAIGtwyobjRecord` object
`Metadata`	metadata__c	Text	Additional information about the source. Not populated for all features. For example, it can be `NOT_SET` for features like Service Replies and Work Summaries.
`Name`	name__c	Text	Name of the Salesforce object. For citations, this is the name of the source in your org. Example values: Article name, case label Not populated for all features. For example, it can be `NOT_SET` for features like Service Replies and Work Summaries.
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Parent`	`parent__c`	Text	Foreign key to the `GenAIGatewayRequest` object or `GenAIFeedback` object
`Record Id`	recordid__c	Text	Record ID of the Salesforce object. For citations, this value represents the record Id of the source int your org.
`Timestamp`	`timestamp__c`	Datetime	Timestamp of when the record was created
Type	type__c	Text	Type of Salesforce Object. For citations, this represents the type of source. Example values: Knowledge article, Salesforce record, Web page
Value	value__c	Text	URL or links to the source Example values: Links to Salesforce objects or URLs to web pages. Not populated for all features. For example, it can be `NOT_SET` for features like Service Replies and Work Summaries.

Note Field population varies by feature. The name__c, value__c, and metadata__c fields aren’t populated for all AI features. For features like Service Replies and Work Summaries, and for app types including ReplyRecommendations and EinsteinGPT, these fields can appear as NOT_SET. This is expected behavior. Only recordId__c and type__c are required for the object context join.

To inspect the full object context for a request, query the GenAIGatewayRequestTag DMO for the same gatewayRequestId__c, filtering for tag key = "entities". The tagValue__c field contains a JSON representation of the GenAIGtwyObjRecord record, including any available context fields.

Einstein Generative AI Gateway Object Record Citation Reference DMO

A Data 360 DMO that captures additional information about the inline citations for the sources used to generate agent responses. The corresponding standard DMO is GenAiGatewayReqObjRecCtn (developer name: GenAiGatewayReqObjRecCtnDmo).

This DMO has these attributes.

Field Label	Field API Name	Data Type	Description
`Cloud`	`cloud__c`	Text	Cost cloud ID. Example values: Service, Sales
`Feature`	`feature__c`	Text	The AI feature associated with the gateway request. Example values: Service Replies, Sales Summaries
`Id`	`id__c`	Text	Primary key: The unique ID to the `GenAIGtwyobjRecord` object.
`Org Id`	`orgId__c`	Text	Salesforce org ID.
`Parent`	`parent__c`	Text	Foreign key to the `GenAIGtwyObjRec` object.
`References`	references__c	Text	Additional information about the citation is presented in a JSON format. Includes information such as: Claim: Explanation of the relevance of the citation to the generated response. Location: A list of locations where the links appear within the response.
`Timestamp`	`timestamp__c`	Datetime	Record creation timestamp.

See Also