🛠️ How-to guide

<aside> 📊

WHAT IT DOES

The model usage report shows exactly what your agents spend on the AI models behind them — total tokens for a period, whether that spend is live customer traffic or your own building and testing, and a full breakdown by activity, agent, model, channel, and team member. Export any view to a spreadsheet.

</aside>

Opening it

From Models, select View usage. Use Back to models to return. Anyone in the workspace can view usage.

Tokens this period

The headline is your total tokens for the selected period, the change vs. the prior period, a Forecast at this rate — where your usage is on track to finish the period — and the number of calls behind it.

Input, output and embedding

Three figures that always add up to the total:

• Input tokens — what your agents send to the model (the question plus its context).
• Output tokens — what the model writes back.
• Embedding tokens — turning your content into searchable meaning for knowledge lookups. These are priced far below chat input, so they're shown on their own.

Where the spend goes

The spend split answers the first question a bill raises — is this real customer traffic, or my own work? Three cards, and selecting one filters the whole report to it:

• Live production — real people talking to your published agents.
• Building & testing — designing agents, testing them, and running evals.
• Workspace & platform — knowledge embedding, search, and Insights widgets.

If some spend can't be attributed yet, an Uncategorised bucket appears — normally empty.

The model usage report: total tokens for the period with a forecast, the spend split (Live production vs. Building & testing vs. Workspace & platform, each a filter), input/output/embedding tokens, and the daily trend with a dashed forecast tail.

The daily trend

A per-day chart with a dashed forecast for the days ahead. Switch what it stacks: