Analyze Cost and Usage
Redpanda Agentic Data Plane prices every LLM call that the AI Gateway routes. Use the Cost and usage report to compare what you spend with the period before. Break the total down to find which agent, model, provider, user, or tag drives it, then export the numbers for reporting.
After reading this page, you will be able to:
-
Read gateway spend, requests, and tokens for a time window against the period before it
-
Break down spend by agent, model, provider, user, or tag and chart one resource against the gateway total
-
Export the breakdown table or the full spending report as CSV
Prerequisites
The report needs traffic through at least one LLM provider. With a provider but no traffic in the window, the figures read zero.
| Dollar values use the per-model rates of the provider that served the call. Those are the provider’s built-in catalog rates or the overrides you set on it. Agentic Data Plane prices a model offered by more than one provider separately for each. Cost reporting works without any pricing setup. See Override per-model pricing. |
Open the report
Open Cost and usage in the sidebar. The page has two tabs:
-
Cost & usage: The spending report for one time window, with CSV export.
-
Activity: The agents active right now, with a drill-down into their recent conversations.
Figures are in US dollars (USD) and cover model costs only. The report groups them by day in UTC, or by hour for a window of 24 hours or less.
Choose the time window
The date-range picker offers last 7 days (the default), last 14 days, last 30 days, last 90 days, month to date, quarter to date, year to date, and a custom range. Every figure on the tab, including each comparison with the previous period, follows this one window.
If the default range holds no traffic, the report switches to the smallest wider range that does, up to the last 90 days. The picker always shows the range in use.
To share what you’re looking at, copy the page URL. It records the range, including a custom start and end time in UTC, so a teammate who opens it sees the same window.
Read the headline numbers
Four figures summarize all gateway traffic in the window:
-
Total spend shows what the window cost, with its change against the previous period.
-
Model requests counts the individual calls to a model.
-
Cost / 1k requests averages spend per thousand requests across all models.
-
Tokens counts input, output, and cached tokens.
The previous period is the window of the same length that ends where yours starts, such as the seven days before a seven-day range. Total spend, the gateway spend chart, and the breakdown table all compare against it. When the previous period had no spend, Total spend shows no percentage change.
A client can call a provider with its own subscription credentials. When the provider reports that the subscription covered the call, the call counts in Model requests and Tokens but adds nothing to Total spend. See Calls covered by a subscription.
Compare a resource against the gateway
To see how much one agent, model, provider, or user contributes to spend over time, chart it next to the gateway total. The gateway spend chart plots spend for all gateway traffic. The Resource spend chart plots one resource on the same dollar scale, so its share of the total reads straight off the axis.
To chart a resource, click Choose a resource on the Resource spend chart and pick a type, then the resource. You can also click Compare trend on a row of the breakdown table. The page URL records the charted resource, so you can share the view. You can’t chart a tag value here, but its row details include a chart of its own spend.
Charting a model covers its whole family, including its aliases and dated versions, so gpt-4o includes gpt-4o-2024-08-06. A name that differs by more than a trailing date is a separate model. For example, gpt-4o-mini is separate, and so is each regional version of a model served through Amazon Bedrock. Charting a model from the Models tab keeps to that row’s provider. Because a row shows the exact model identifier the gateway recorded, a charted model’s spend can exceed the row you charted it from.
|
Break down spend
Use the Spend breakdown table to find what drives spend. Group it by agent, model, provider, or user with the table’s Agents, Models, Providers, and Users tabs. When cost-allocation tags are available, a Tags tab groups spend by the values of the tag key you choose. Use it to split a shared provider’s bill by team or department. See Track Spend by Tag.
Grouping applies to this table only. The headline figures and the gateway spend chart always cover all gateway traffic, and the Resource spend chart keeps whatever you charted.
On the Models tab, each row is one model as served by one provider. A model you reach through two providers gets two rows, so you can compare what each route costs.
Some rows group spend that has no resource of that kind. Direct usage is calls that didn’t run through an agent. Unattributed is calls with no user recorded. Untagged is calls without a value for the chosen tag key. A deleted provider keeps its row, marked Deleted provider, so history stays complete. If the window holds more resources than the table can load, the table says so. Narrow the range, or export the Full report, every row (CSV), which isn’t truncated. See Export the report as CSV.
Click a row to see its share of gateway spend, its failed requests, and its token mix, or to open its agent or provider page.
Export the report as CSV
The Export menu offers two CSV downloads:
-
Breakdown table as shown (CSV): The breakdown table with its current grouping, search, and sort order.
-
Full report, every row (CSV): The complete dataset for the window, across all gateway traffic.
In the same menu, set the full report’s time bucket (hourly, daily, monthly, or one total). Then choose which of provider, model, user, and agent group its rows.
Use the breakdown export to share the view you have on screen, and the full report when you need the whole dataset for a spreadsheet or a chargeback pipeline.
See which agents are active now
The Activity tab answers a different question from the report. Instead of what something cost, it shows what’s running right now. It ranks the agents with traffic over the last hour (the default), the last 24 hours, or the last seven days. Each agent shows its change against the previous window. Select an agent to see its recent requests and open a conversation.
Each conversation shows the estimated cost of the model calls it made through AI Gateway, priced the same way budgets count them. Use it to find the conversations that drive an agent’s spend. A total with an asterisk covers only the calls AI Gateway could price. A conversation reads Not priced when AI Gateway priced none of its calls, or when Agentic Data Plane couldn’t read its cost.