Considerations for Agentforce Service Agent

Your data isn’t retained by third-party LLMs. We partner with all our external model providers, such as OpenAI, to enforce the zero-data retention policy.

• No data is used for LLM model training or product improvements by third-party LLMs.
• No data is retained by the third-party LLMs.
• No human being at the third-party provider looks at data sent to their LLM unless Salesforce grants view access in response to documented suspicion of illegal activity or significant violation of applicable usage policies.

• Relevant information from a Salesforce record is merged with the prompt to provide context.
• Secure data retrieval of Salesforce data for grounding the prompt based on the permissions of the user executing the prompt.
• Secure data retrieval preserves in place all standard Salesforce role-based controls for user permissions and field-level security when merging grounding data from your CRM instance or Data 360.

• System policies help limit hallucinations and decrease the likelihood of unintended or harmful outputs by the LLM.
• System policies can vary for different generative AI features and use cases.

• Prompts, responses, and trust signals are logged and stored in Data 360.
• Feedback can be used for improving prompt templates.
• Provides pre-built reports and dashboards for analysis.

Usage is determined by the number of standard actions called by the agent in a text conversation. Currently, every standard action called by the agent is metered as one standard action, irrespective of the number of tokens processed.

Standard agent actions are actions that are available out-of-the-box. To check the list of Standard actions, see Standard Action Reference.

Note Use of some standard agent actions require that a subscription has been purchased for each user that accesses these actions, such as a subscription to Agentforce for Sales Add-on or Agentforce for Service Add-on. To determine which subscription is required for such standard actions, see Standard Action Reference at Standard Action Reference. While this requirement is not technically enforced yet, users who don’t have the required add-on license will lose access to such actions for which they don’t have a license when the requirement is enforced.

Usage is determined by the number of custom actions called by the agent in a text conversation. Currently, every custom action called by the agent is metered as one custom action, irrespective of the number of tokens processed.

Async calls to Salesforce AI in custom action implementation may incur additional charges.

Usage is determined by the number of standard actions executed by the agent in a voice conversation.

Note Use of some standard agent actions require that a subscription has been purchased for each user that accesses these actions, such as a subscription to Agentforce for Sales Add-on or Agentforce for Service Add-on. To determine which subscription is required for such standard actions, see Standard Action Reference at Standard Action Reference. While this requirement is not technically enforced yet, users who don’t have the required add-on license will lose access to such actions for which they don’t have a license when the requirement is enforced.

Usage is determined by the number of custom actions called by the agent in a voice conversation.

Custom actions are actions that are created by you or which result from your modification of a standard action. To learn more about what customer created actions are, see Create a Custom Agent Action. To learn more about how standard actions become custom actions, see Editing Standard Agent Action Reference Actions.

Async calls to Salesforce AI in custom action implementation may incur additional charges.

For this subtype of the Agentforce usage type, usage is calculated based on conversation windows. A conversation window starts:

• when the user sends a message for the first time after the previous conversation window ends, or
• when ASA sends a pre-configured welcome message first.

The end of the conversation window differs for each Digital Engagement channel on which ASA is enabled.

For the Enhanced Chat channels:

• Enhanced Web Chat: For unauthenticated users, the conversation window ends when the user explicitly clicks End Conversation. For authenticated users, the conversation window ends 24 hours after the start of the conversation.
• Enhanced In-App Chat: The conversation window ends 24 hours after the start of the conversation.

For these channels, the conversation window ends 24 hours after the start of the conversation:

• Enhanced WhatsApp
• Enhanced Facebook Messenger
• Enhanced Apple Messages for Business
• Enhanced SMS
• Bring Your Own Channel
• Enhanced LINE

Agentforce Service Agent (ASA) only supports inbound conversations, which are conversations that a user initiates. See Generative AI Billable Usage Types.

Einstein Requests is one of the usage metrics for generative AI. The use of generative AI capabilities, in either a production or sandbox environment, can consume Einstein Requests and possibly Data 360 credits. To learn when Einstein Requests applies, see Metering for Agentforce and Generative AI Usage.

Direct API calls to the LLM gateway use Einstein Requests. To learn more, see the Rate Card for Einstein Requests.

Einstein Requests are available in:

• Enterprise, Performance, and Unlimited editions with an Einstein for Sales, Einstein for Platform, or Einstein for Service add-on. To purchase the Einstein for Sales, Einstein for Platform, or Einstein for Service add-on, contact your Salesforce account executive.
• All Einstein 1 Editions
• Salesforce Enterprise and Unlimited Editions with Marketing Cloud Next Growth or Advanced Edition

Usage is calculated based on the number of records processed.

The count of records processed depends on the structure of a query as well as other related factors such as the total number of records in the objects being queried.

Usage is calculated based on the amount of unstructured data that is processed without using AI assisted content processing and extraction. For example, if the search index processes 100 PDF documents that are 1 MB each, usage is calculated as 100 MB. If the search index processes five audio/video files that are on average 100MB each, usage is calculated as 500MB.

In Data 360, unstructured data may be chunked and vectorized using an embedding model. Usage is computed only once across both these activities. For example, if one 100 MB PDF document is chunked and vectorized, usage is computed as 100 MB, not as 200MB.

If you create a search index that uses a data model object (DMO) as a data source and select Text data type fields from the DMO, all strings from those fields are indexed and the sum total of all bytes processed is reported as Unstructured Data Processed.