AI Engine records each AI action in the AI Activity Log object. The log gives admins and auditors a trail of who used which feature, on which record, with which model, and whether it worked. It holds usage metadata, plus a short error message when a call fails. It does not hold the full prompt or answer text.
What each log record contains#
One record is written per AI action. An action that makes several provider calls (for example a chat question: one call to write the query, one to write the answer) is one record, with the number of calls in Callouts.
| Field | Content |
|---|---|
| User | The user who ran the action. |
| Action Type | The feature: Chat, Summary, Next Steps, Email Draft, Web Check, Company News, Email Analysis, Meeting Prep, PDF, Custom Prompt, Custom Prompt (Global), Scoring, Magic Snap, User Help or Admin Help. |
| Object Type and Record Id | The object and record the action ran on, where there is one. |
| Model and Models Used | The model used, and a per-model token breakdown when one action used more than one model. |
| Callouts | Number of requests sent to the provider for this action. |
| Tokens Input, Tokens Output, Tokens Total | Token counts as reported by the provider. |
| Tokens Cache Read, Tokens Cache Write | Input tokens served from or written to the provider's prompt cache. |
| Web Searches | Number of web searches the provider ran. |
| Duration ms | How long the action took. |
| Estimated Cost USD | An estimate; see Token usage and cost. |
| Success | Whether the action completed. |
| Error Message | When a call fails, the error message, clipped to 500 characters. A provider's error message can quote part of the request. The field can also carry short status notes, such as how many data-source rows a prompt used. |
What the log does not contain#
AI Engine does not keep a copy of the full request sent to the provider or of the full answer, in the log or elsewhere. The log shows that an action happened and what it cost, not the conversation itself. AI-generated text is saved in Salesforce in two cases: when a user chooses to save it, for example by creating a task, sending an email, saving a PDF to Files or saving a Magic Snap record, and when AI Scoring writes a score, label and written reason to the score fields on a Lead or Opportunity. See How AI Engine handles your data.
Who can see the log#
- Admins with AI Engine Admin can see and delete all log records, and use the Activity Monitor.
- Users with AI Engine User have read access to the object. With the org-wide default set to Private, they can read only their own records (plus, while Grant Access Using Hierarchies is on for the object, those of users below them in the role hierarchy).
- Users do not need create or edit access to the log. AI Engine writes the records itself, and each record is owned by the user who ran the action.
Clear Grant Access Using Hierarchies for the object in Sharing Settings if managers should not see their team's log records. We recommend keeping the AI Activity Log org-wide default at Private and checking it after upgrades; see Recommended security settings.
Review activity in the Activity Monitor#
The Activity Monitor tab in the admin panel (also available as the aiActivityMonitor component) summarises the log: calls, tokens and estimated cost, usage by feature and user, and the 200 most recent log records with their status. Only users with AI Engine Admin can load its data. See Activity Monitor.
Using the log for audits#
- Failures: look for rows marked ✗ Error in Recent Logs, or check the Success field on exported records, to find configuration or provider problems.
- Access patterns: check which users use which features and on which objects. A spike for one user or one feature is easy to spot in the monthly views.
- Record-level questions: the Record Id field on each log record shows which records a user ran AI actions on. Recent Logs does not show it, so use a report or an export.
The AI Activity Log object has no tab of its own, and the package does not enable reports on it. To report on it, tick Allow Reports on the object and create a custom report type once, as described in Activity Monitor. To work with the records outside Salesforce, export them, for example with Data Loader.
Retention#
AI Engine does not delete log records automatically; they stay until someone deletes them. If your policy requires a retention period, delete older records on a schedule with your usual Salesforce tools. To keep a long-term archive, export the records first, for example with Data Loader or the Salesforce data export.