You are here:
Metering for Agentforce and Generative AI Usage
Generative AI usage for Agentforce features is typically metered and consumes credits.In some cases, when required conditions are met, usage can be unmetered and doesn’t consume credits.
Unmetered Usage
Usage is unmetered when it meets specific conditions such as to user type, assigned permissions, and the execution context. There are two categories of unmetering.
- Global Unmetering: Product-agnostic logic based on whether AI usage meets general organizational and user-level conditions.
- Product-Based Unmetering: Applies to a specific list of products when used with designated licenses and permissions.
Key Conditions for Global Unmetering an Agentforce Action or AI feature
- The user account must be assigned to a System Administrator or Standard User profile.
- The user must have the Unmetered User-Based AI permission.
- The action or AI feature use must be executed as the current user with the above unmetered permissions, and not as an automated process user or any other identity.
- Unmetered usage is limited to a maximum of 30 LLM calls per minute per user. This limit is not currently technically enforced in the product.
Unmetering Logic for Global Unmetering
This table lists the conditions and the type of metering for Agentforce features.
| Type of AI usage | Agent Type | User Type | unmetered User-based AI permission | Agent Action or Prompt Run as current user | metering behavior |
|---|---|---|---|---|---|
| Features that are never metered. For example, certain Trust and Guardrail features. Note:Using additional services, including Data 360, can consume credits. | Not applicable | Not applicable | Not applicable | Not applicable | Unmetered |
| Prompt-based, non-agentic AI usage. For example, a flow running a prompt template. | Not applicable | Standard or Admin | Yes | Yes | Unmetered |
| Agentforce Agent Action | Employee Agent | Standard or Admin | Yes | Yes | Unmetered |
| Agentforce Agent Action | Sales Coach | Standard or Admin | Yes | Yes | Unmetered |
Examples of Metered and Unmetered Scenarios
| Example | Metering status | Reason |
|---|---|---|
| A user clicks a button in the UI to launch a prompt template. | Unmetered | Executed as current user with appropriate permissions. |
| A Flow launches from a record update and runs as the Automated Process User. | Metered | Execution context is not the current user. |
| A Flow launches from a record update and runs as the current user. | Unmetered | Executed as current user with appropriate permissions. |
| Scheduled Apex jobs or batch processes running in the background. | Metered | Automated execution without direct user interaction. |
| An Agentforce agent runs on a schedule to summarize cases. | Metered | Scheduled execution context. |
Product-Based Unmetering
Generally, when a logged-in user with appropriate permissions interacts with an AI feature or agent, usage of certain features can be unmetered. Product-based unmetering is currently available only with a limited set of Agentforce Add-ons.
- Only specific Agentforce and embedded AI features are unmetered, excluding them from consuming Flex, Conversation, or Einstein Requests. If AI usage triggers underlying services such as Data 360, other credits including Flex Credits are still consumed for such underlying services.
- Unmetered usage is limited to a maximum of 30 Large Language Model (LLM) calls per minute per user. This usage limit isn’t currently technically enforced within the product interface.
Metered Usage
Usage that does not meet the unmetering conditions consumes either Flex Credits or Einstein Requests. If AI usage triggers services like Data 360, those specific credits are still consumed even if the AI call itself is unmetered.
