Skip to navigation
Settings

Usage data

Use Usage to review flow runs, AI token usage, and Runtype-billed costs. In an organization, Usage shows organization-wide usage, including your teammates’ runs. In a personal account, it shows your own usage.

View usage data

To review usage analytics, follow these steps:

  1. In the sidebar, click Settings > Usage. The page opens with Billing month selected.
  2. To change the reporting period, select 7d, 30d, or 90d in Period.
  3. To reload the data immediately, click Refresh usage analytics (the refresh button next to Export CSV).
  4. To download the report, click Export CSV. The CSV file contains data for the selected period.

The page refreshes automatically every 60 seconds. The Usage page includes only activity billed by Runtype. Charges from your own API keys don’t appear in this report.

Understand usage metrics

When the selected period contains usage, the Usage page starts with summary tiles. It shows Runtype-billed usage, Executions, and Average cost per execution. When cache activity exists, it also shows Estimated cache savings and cache hit rate. If no usage data exists, the page shows No usage data yet.

When usage data exists, the page shows the following charts and tables:

  • Usage cost over time: Shows daily Runtype-billed cost.
  • Executions over time: Shows execution volume by day.
  • Usage cost by category: Splits cost between AI models and built-in tools.
  • Cost by AI model: Lists models from highest cost to lowest with request count, token count, cost, and cost per request. It also shows cached input share when available.
  • Cost by built-in tool: Lists requests and cost for each built-in tool.
  • Prompt cache: Shows cache read and write token totals. The page displays this card only when the selected period contains cache activity. For more information, see Prompt caching.

The totals, charts, breakdowns, and CSV export use the selected period.

For this report, an execution is one flow run, batch record run, eval run, or scheduled trigger. One execution can make multiple model or tool requests.

When data loads, the page displays when the report was last updated. If the selected period starts before the available history, the page warns that earlier dates might be unavailable. Treat missing coverage as unknown, not as zero usage.

Usage compared with billing

Recent activity can appear in Settings > Billing before it appears on the Usage page. To review your spend toward your monthly cap, or to view invoices, use Settings > Billing. For more information, see Billing and plans.

Read the CSV export

The exported file is named runtype-usage-SCOPE-DATE.csv, where SCOPE is organization or personal and DATE is the date that the report was generated. The file has the following columns:

ColumnContains
record_typeThe row type: summary, daily_usage, daily_executions, model, or tool
date_or_periodThe day for daily rows, or the date range for the summary row
itemThe model name, the built-in tool name, or a label for totals
requestsThe request count for model and tool rows
executionsThe execution count for summary and daily_executions rows
tokensThe token count for model rows
usage_cost_usdThe Runtype-billed cost in US dollars, to five decimal places

Cells that don’t apply to a row type are empty.

Investigate contributing executions

Click Runtype-billed usage, Executions, or a model or tool name to open the executions behind that selection. In end-user reporting, click a user’s spend or execution count. The investigation keeps the selected dates and filters.

Choose an execution to open its trace. Select a trace event to inspect its details and use Add to eval to turn a useful example into a regression case.

The execution list covers the last 93 days. Older usage, and usage that isn’t linked to a specific execution, doesn’t appear in the list, so an empty list doesn’t mean that there was no usage. Totals in this list can differ slightly from the overview charts. Runtype-billed cost includes pricing adjustments, so it can differ from the cost estimates shown in a trace.

Reduce usage costs

Use these methods to lower your Runtype-billed costs:

  • Use a lower-cost model. For routine tasks, choose a model such as gpt-5.6-luna instead of gpt-5.6-sol.
  • Set a lower output limit. When a shorter response is enough, set a lower maxTokens value. This caps the output tokens for each response.
  • Store reusable API results. Save API results in Records so that later runs read them instead of calling the API again.
  • Shorten prompts. Remove prompt content that the model doesn’t need. Fewer input tokens cost less.
  • Reuse cached prompt content. For more information, see Prompt caching.

Understand bring your own key (BYOK) charges

When you use your own API keys (BYOK), your providers bill you directly for model costs. Those charges don’t appear in the cost totals, charts, or model breakdown on the Usage page. The runs can still appear in the execution counts. To see exact charges, check each provider’s billing dashboard. Treat Runtype cost estimates as planning data, not invoices.

Troubleshoot usage data

If the usage data doesn’t look right, use the following fixes:

  • The page shows Failed to load usage data. Click Retry.
  • The page shows No usage data yet. Check that the selected period contains activity, or select a longer period.
  • Costs are lower than you expect. Model costs from your own API keys don’t appear in the cost totals. Check each provider’s billing dashboard for those charges.
  • The execution list shows Could not load contributing executions. Click Retry. The error doesn’t change your usage totals.

Next steps