Concepts
Audit trail
The audit trail is the record of every AI request that passes through msharti.
What each entry records
- Who asked, and which agent acted for them.
- The connector, tool and action.
- The role check result (allow or deny).
- The approval decision and approver, where needed.
- The result and the time.
Example entry
// Worked
{
"time": "2026-10-26T09:14:02Z",
"user": "w.kamau",
"agent": "assistant",
"tool": "update_loan_note",
"action": "write",
"role_check": "allow",
"approval": { "status": "approved", "by": "supervisor.otieno" },
"result": "success"
}
Who can see it
Access is limited to the named reviewers agreed with your organisation, such as your risk owner or administrator.
Retention and export
Retention, export formats and integrity protections are documented in your agreement and data processing agreement (DPA).
Privacy
The audit fields and content-handling rules are agreed for your deployment. The aim is to record the decision and result while minimising personal data.
Policy checked · role=finance
Reconciliation · MPE-S06•••
Decision and result linkedrequest → policy → approval → action
- 09:14Request received
- 09:15Approval recorded
- 09:17Result logged
Who asked. Who approved. What changed.