---
title: Usage analytics
description: Monitor inference volume, inspect cache-aware token breakdowns, and track financial metrics in the Zorveus dashboard.
---

The [Zorveus Dashboard](https://app.zorveus.com) provides real-time visibility into every inference request processed across your organization.

Navigate to **Usage** in the dashboard to analyze token volume, prompt cache savings, and cost metrics.

---

## Filtering and exploring usage

At the top of the **Usage** page, you can slice and filter metrics using:

- **Date range**: Presets (Today, Last 7 Days, Month to Date) or custom calendar ranges.
- **Application**: Filter by specific app connections or view aggregate organization traffic.
- **Model and provider**: Group by provider (OpenAI, Anthropic, Google) or isolate specific models.
- **Billing mode**: Filter between `wallet` and `byok` traffic.
- **Product user**: Search by `external_user_id` or customer name to review one customer's consumption.

---

## The usage events log

Below the aggregate summary charts, the **Usage events** table logs individual completed requests with their key metrics:

| Column | Description |
| :--- | :--- |
| **Timestamp** | UTC completion time. |
| **Model** | Resolved model identifier (such as `anthropic/claude-sonnet-4-5`). |
| **Input tokens** | Total input tokens. If prompt caching occurred, cache reads appear as a badge under the count. |
| **Output tokens** | Generated completion tokens. |
| **Usage cost** | Reference-priced virtual spend consumed by the call. |
| **Billed to wallet** | Actual cash charge debited from your organization wallet ($0.00 for BYOK unless a platform fee applied). |
| **Mode** | `wallet` or `byok`. |
| **Status** | `succeeded`, `partially_settled`, or `failed`. |

---

## Inspecting prompt cache savings

When an upstream model provider reports prompt caching, Zorveus highlights the savings in the dashboard.

In the usage table, cache reads appear directly under total input:

```text
Input
10,000
7,000 cache read
```

Clicking any row in the table opens the **Event details** drawer with an **Input details** breakdown:

```text
Total input                         10,000
Normal input                         1,000
Cache read                           7,000
Cache creation, default duration     1,500
Cache creation, one hour               500
Output                               2,000
Service tier                      standard
AI allowance saved                 $0.0105
```

### What the cache metrics mean

- **Normal input**: Uncached prompt tokens calculated as total input minus cache reads and cache creations.
- **Cache read**: Tokens reused from an active prompt cache at discounted provider rates.
- **Cache creation**: New tokens written to cache for subsequent calls.
- **AI allowance saved**: Reference-priced allowance saved by prompt caching compared to standard input rates. On BYOK requests, this represents virtual allowance saved rather than discounts on your provider invoice.

---

## Understanding financial amounts in the dashboard

The event detail drawer separates three distinct financial values:

1. **Virtual spend ("Usage cost")**: The reference-priced value of AI usage evaluated against spending caps and user credits. Wallet and BYOK calls compute this value identically.
2. **Sell cost ("Billed to wallet")**: The actual amount deducted from your organization wallet balance.
3. **Provider cost**: The raw wholesale amount billed by the model provider.

If a call exceeded its allowance under an overrun policy, the drawer also displays:

- **Uncovered virtual spend**: The portion of spend exceeding base cap and credits.
- **Overrun policy**: The policy active for the request (`limit_output`, `allow_overrun`, or `carry_overrun_forward`).