# Audit activity and understand usage

Use the Audit log to inspect actions and Usage to understand resource consumption. They report different evidence about your agent’s work.

## Audit log
Open **Settings → Audit log**. Filter by outcome, caller, downstream service, and date range. Inspect tool calls, approvals, denials, and redactions relevant to the event you are investigating.

Read the event detail and any integrity result. A failed integrity check means the record did not match its integrity data; do not treat it as a trustworthy account without investigation. An integrity check supports the checked records, not a claim that every external system behaved correctly.

Export the relevant records when you need to review them outside the page. Check the export before sharing and keep credentials out of a support request.

## Usage
Open **Settings → Usage** and choose the period or filters. Review breakdowns by model, source, day, and purpose. Chat, subagents, loops, and automations can all contribute. Cost estimates depend on the pricing and usage recorded; they are not necessarily a final invoice from an external service.

## Read counters correctly
Ledger totals persist differently from the **Since gateway start** live counters, which reset on restart. Cache savings describe the recorded cache behavior; they are not proof that every request avoided cost.

## Control future work
Use Guardrails for unattended budgets. Reducing a budget can prevent later runs; it does not undo spending already incurred. Check skipped jobs if an expected routine stops producing results.