8. Monitor Usage
Once your applications are calling models through AI Gateway, use the Usage page in your workspace to monitor consumption, track costs, and identify failed requests.
Open the Usage Page
In your workspace, open Usage. The page displays token consumption, estimated costs, and error metrics for the selected workspace.
Filter the Data
Use the filters at the top of the page to narrow the displayed data:
| Filter | Description |
|---|---|
| Time Range | Select the period to analyze, such as Last 24 Hours. |
| Models | View usage across all models or focus on a specific model. |
| Credentials | View data for all credentials or a specific credential to identify which application is generating the most usage. |
Summary Cards
The summary cards provide aggregated metrics based on the selected filters.
| Card | Description |
|---|---|
| Requests | Total number of API requests made. |
| Total Tokens | Total input and output tokens consumed. |
| Tokens (In / Out) | Comparison of input tokens (prompts) and output tokens (model responses). |
| Estimated Cost | Estimated spend for the selected period. Actual charges are finalized during billing. |
| Average Cost per Request | Estimated cost divided by the total number of requests. |
| Successful | Number of requests completed successfully. |
| Cache-Read Tokens | Input tokens served from the model cache instead of being processed again. |
| Cache-Write Tokens | Input tokens stored in the cache for reuse by future requests. |
| Reasoning Tokens | Tokens consumed internally by reasoning models while generating responses. |
A newly created workspace displays
0for all metrics until requests are made through AI Gateway.
Charts
The charts section provides a visual breakdown of usage over time:
- Requests by Model: Displays the number of requests handled by each model during the selected period, helping identify the most frequently used models.
- Token Volume: Shows token-consumption trends over time, making it easier to identify usage spikes and patterns.
- Errors: Displays failed requests over time, enabling early detection and investigation of issues.
Export
At the bottom of the page, a detailed request log is available in tabular format. You can filter the data by attributes such as model, credential, and user to analyze usage patterns in greater detail.
To download the data for offline analysis or reporting, click Export CSV on the right side of the table.
Tips
- Select a specific credential to determine which application is driving the most usage or cost.
- If the Errors chart shows an increase in failures, review the corresponding status codes in the Common Errors section of /docs/first-api-call.