Skip to main content
Settings → Usage shows recorded token usage by day, project, and member. Tokens by work origin compares Self-initiated, Requested, and Unknown token totals for the same selected scope and period.

​
What is counted

The headline, chart, By project, and By member count turns from Cogno’s background work queue. The separate origin comparison also includes measured Task-linked input through Cogno chat on personal machines. It counts measured executions attributable to the workspace; background project runs without a Task remain Unknown. The comparison shows recorded token totals, not the additional tokens caused by adopting Cogno. Runs on a Workspace’s EC2 Engine are attributed to the member recorded when the run started, including delayed reports after a restart. Other Managed Cloud runs, CLI usage outside Cogno, and chat without a Task are not included. The Task’s author does not determine the classification. A turn whose executor reported nothing is not counted as zero; it is simply absent. Claude Code and Codex both report token counts, so this is rare.

​
Read the page

1

Open Settings → Usage

The page opens on You and Last 7 days.
2

Choose whose usage to see

You includes your own machines and your runs on the Workspace’s EC2 Engine. Workspace is everyone’s, and only the workspace owner can open it.
3

Choose a period

Last 7 days, Last 30 days or Last 90 days. The bars widen with the span — a day each for a week or a month, a week each for ninety days — so the chart stays readable. The label under the chart says which.
The headline is the total tokens, followed by the number of turns behind it and the split between input, cache read, cache write and output. The split matters: a cached read costs a fraction of a fresh one, so two workspaces with the same total can be spending very differently. Below the chart, By project and By member rank where the work went. Long lists are folded: the top eight are named and the rest arrive as a single row saying how many are behind it.

​
Cost

Claude Code reports a cost with each turn and the page shows it, described as a list price. It is not an amount anyone was billed — a subscription plan charges none of it. Codex reports no price at all. A Codex-only period therefore shows tokens and no cost, which is different from a cost of zero.

​
Where the numbers come from

Each measured turn reports what it spent to Cogno as it ends, and the figures are the executor’s own — the same ones claude and codex count for themselves. Nothing is estimated here. Usage is kept indefinitely. The period control limits what one view asks for, not how long the record lasts.

​
Read work time and proposal decisions

Open Work time to review completed Task runs’ AI execution time. Choose Self-started / requested to split the time by recorded input. Human Task messages count as requested. Initial runs use the Task’s recorded mention or ambient conversation source. Inputs whose origin cannot be verified, including agent follow-ups, remain unknown. A Task’s author does not determine this split. The adoption figure uses each new-Task proposal’s latest explicit human decision in the last four weeks: adopted divided by adopted plus declined. For Cogno-authored Tasks with an ambient source, a person’s move to Ready, In Progress, or Done counts as adoption; a move to Rejected counts as declined. Archived Tasks, unchanged ordering, automatic changes, and changes cascaded from a parent do not supply a decision. Undo takes the status choice back. Explicit decisions from the older proposal controls also count once per Task. Unanswered proposals and later agent status-change proposals are excluded. When no proposal has an explicit decision, the rate displays -. The four-week window does not change with the chart period.
  • Machines — the computers that run autonomous turns
  • Agents — what Cogno does unattended
  • Billing — the plan and what it actually costs