# Usage

Review committed AI-credit usage by date and agent.

**Usage** shows committed AI credits for the active billing owner. Choose a date range and optionally select one agent.

The dashboard includes:

- Total committed credits in the selected period compared with the monthly limit
- A breakdown of response, successful-action, and attachment credits
- The number of successful action calls
- A daily usage chart
- A per-agent breakdown of committed credits

## How chat credits are calculated

Let **C** be the selected model's credit rate:

```text
Total turn credits = response C + (successful action calls × C)
```

A response without an action uses **C**. A response with two successful actions uses **3C**. Configured actions and successful Helpdesk handoff calls each add action credits; Knowledge Base searches do not.

Each successful `requestHumanSupport` invocation is one action call. When Helpdesk first asks for an email and then creates the ticket on a later turn, each successful invocation uses the selected model's credit rate.

Action credits are reserved before execution and committed only for a successful result. Failed actions release their reservations. If an action cannot reserve its credits, it is not run. The AI response can still use its normal response credits when it reports that outcome.

Showing a configured button or form is the billable action. A visitor later clicking that button or submitting that already-rendered form is not another AI action and uses no additional credits.

Only committed AI responses, successful action calls, and attachment processing appear as used credits. Reservations refunded after a failed or aborted operation are not treated as committed usage.

Usage is an operational view, not an invoice or payment ledger. Open [Billing](/docs/workspace-and-billing/billing) for plan, renewal, purchase, and transaction information.

## Investigate unexpected usage

Narrow the date range, select the agent responsible for the increase, and compare the result with that agent's Chat Logs. Model choice sets both the response rate and the per-action rate, so review recent model changes, successful action calls, attachment processing, and conversation volume.