Analytics
The Usage Analytics page shows usage data across every agent and every end user on your deployment. Navigate to it via Analytics in the navigation drawer.
Use it to answer questions like: "Which agent is the most used?", "Which model is driving cost?", and "How does activity compare week over week?"
Time range
Three preset buttons in the top-right toggle the time range:
- 24H — last 24 hours.
- 7D — last 7 days.
- 30D — last 30 days.
A Refresh button beside the toggle reloads the data manually.
Summary cards
Three cards at the top summarise the selected time window:
| Card | What it shows |
|---|---|
| Total Normalized Cost | Estimated cost normalized across model pricing tiers. |
| Total Actual Cost | Raw dollar amount based on token usage and model pricing. |
| Total Requests | Total number of conversation requests in the window. |
Costs are estimates derived from the model catalog's per-million-token pricing. They are a useful proxy for understanding spend trends — not an invoice. For total deployment cost (including AWS infrastructure), cross-check with AWS Cost Explorer.
Usage breakdown sections
Below the summary cards, three sections break down usage from different angles:
| Section | Groups by | Useful for |
|---|---|---|
| Usage by User | End-user ID | Spotting heavy users; per-user quota decisions. |
| Usage by Agent | Agent name | Finding which agents are doing the most work. |
| Usage by Model | Model ID (provider:model-id) | Identifying which models drive cost. |
When there is data for the selected time window, each section shows sortable tables and charts. When there is no activity, sections display "No data available for this grouping".
Notes
- Numbers are computed from usage events the gateway records on every turn. There can be a small lag (seconds to a minute) between a chat ending and the analytics reflecting it.
- The Total Normalized Cost and Total Actual Cost figures reflect model token pricing only. They do not include AWS infrastructure charges (Bedrock, ECS Fargate, DynamoDB, OpenSearch Serverless, etc.) — cross-check with AWS Cost Explorer for the full picture.
- Numbers are deployment-wide. End users see their own narrowed view of usage on the Usage & Quota page in the User Dashboard.
Tips
- Check "Usage by Model" first when costs spike. A single agent switched to a frontier model can dominate a deployment's bill within days.
- Compare 7D vs 30D. A spike in 7D that doesn't appear in 30D points to a recent change worth investigating.
- Use alongside AWS Cost Explorer. The analytics page gives you the model-level picture; Cost Explorer gives you the infrastructure picture.