Audit trail (agent actions)
An audit trail is the logged record of every action an AI agent takes: which tool was called, with what arguments, what the result was, and when. A good audit trail lets the user review what the agent did without trusting the agent's narration.
Why it matters
An AI agent that takes real action against real systems will, occasionally, do the wrong thing. The wrong thing might be the agent's mistake (hallucinated tool arguments), the user's mistake (ambiguous instruction), or a tool's mistake (API change broke the contract). When something goes wrong, the question is "what actually happened," not "what did the agent claim happened."
The audit trail is the answer. Every tool call, with its arguments and result, written to a log the user can review. The narration the agent produces in conversation is for human ergonomics; the audit trail is the source of truth.
For most consumer agent products, the audit trail is operationally invisible. Users do not browse it daily. But when something does go wrong (a misdirected email, an unexpected charge, a missed reminder), the audit trail is what lets the user understand and recover. Without one, the only option is to trust the agent's story; with one, the user can verify.
For business agents, audit trails are also a compliance surface. Regulated industries require records of automated actions; a structured audit trail produces them by default.
How Ajeris handles this
Every tool call the agent makes writes to a per-user audit log: the tool name, the arguments, the result (or error), the timestamp, and the channel the conversation was happening on. Users can review the log at any time through the manage page. Approval-gated actions also write to the log when they are authorized or rejected.
The log is per-user isolated; one user's audit cannot reach another's. It is kept with the account and deleted when the account is deleted.
