--- title: 'Usage history' description: 'Compare request, token and spending history by model or provider, and download request metadata.' url: 'https://docs.streamtoken.dev/usage-history' last_updated: '2026-11-11' --- > Docs index: https://docs.streamtoken.dev/llms.txt · MCP: https://docs.streamtoken.dev/mcp · Skill: https://docs.streamtoken.dev/skill.md # Read your usage history Owners and administrators can open **Usage** to inspect organization request metadata. The history chart uses the selected workspace's inference region and the same API-key, department, model, test-traffic and retention filters as the request list. It contains no request or response content. ## Choose a reporting window - **Day** shows 24 UTC hourly intervals on the selected date. - **7d** shows seven UTC dates, ending on the selected date. It is not a calendar-week selector. - **30d** shows 30 UTC dates, ending on the selected date. It is not a calendar month. The visible date interval and region identify the chart's scope. Use the date picker to move the window. Future dates are capped at today; invalid dates use today. Boundaries include the first date's midnight and exclude midnight immediately after the final date. Empty intervals remain visible. The 30-day option applies to Usage and its download. Credits and Diagnostics keep Day and Week. ## Group real activity Choose **By model** or **By provider**. Each slim bar represents an hour or date; its colored segments add up to that interval's selected metric. Provider comes from the model catalogue. An unrecognized provider appears as **Unknown provider**. Metrics are **Requests**, **Recorded tokens**, **Input tokens**, **Output tokens**, **Errors**, **Average latency**, **Wallet charge**, and **Own-key estimate**. Errors count HTTP statuses of 400 or greater. Token metrics sum only the counts actually recorded. **Average latency** is the request duration of all matching requests in each interval, weighted by request count. It is one unstacked bar per interval, because averages cannot be added across models. The legend shows each group's own average instead of a share. The legend shows each group's total and share of the selected metric. When more than five groups exist, the chart displays the four largest and combines the rest into **Other**, with the number of combined models or providers. Other's intervals, total and share include every remaining group. Ranking is based on the selected metric. When the total is zero, a share is unavailable rather than a division by zero. Hover over a chart segment to see its exact group value for that UTC interval. Bar segments are also keyboard focusable and expose the same value to assistive technology. The separate detailed interval table has been removed. Compact axis labels may round values or use scientific notation for very small amounts; tooltips and legend totals retain precision. ## Interpret spending Spending has two separate metrics. They are never added together. - **Wallet charge** is the settled TokenStreamer credit charge for funded requests. Pending holds appear only after they settle. - **Own-key estimate** is the reference-rate estimate for requests on your own provider keys. It is not an invoice or a TokenStreamer charge. It excludes cache adjustments and account-specific discounts. Requests without usable usage/rate metadata have no amount; they are not silently treated as free. Internal platform provider costs are never shown. A recorded zero is distinct from **Not estimated**. Known amounts are added, even if other requests in the interval lack estimates. The estimate coverage count shows how many matching requests have estimates. Legend shares for spending use only the known estimated total. A dashed baseline mark identifies an interval with requests but no estimates; a true empty interval has no mark. No matching requests produces an empty chart and an explicit message. Usage metadata retention applies before chart aggregation. A 30-day chart does not recover records hidden or deleted under a shorter retention policy. Chart totals cover the entire filtered window and do not change when the recent-request list moves between pages. ## Spending amounts For funded requests, the amount shown is the credit charge that settled for the request. For requests on your own provider keys, it is a reference-rate estimate; your provider bills you directly. Missing or unresolved usage stays unpriced. Token Streamer's internal provider costs are never shown. In exports, the combined amount is the `provider_estimate_picos` field. Divide it by 1,000,000,000,000 for US dollars. Each request shows whether it was streamed when the router reported it. Older requests without that information show as unknown. A management API for reading usage from code is planned.