Usage and cost
The usage bar with each harness's subscription windows, the spend of sessions, Workers and groups, and the budget per group.
The app shows two different things about what your agents use:
- Subscription usage: how much of each harness's plan you have used in its current windows, such as 5 hours or weekly, and when they reset. This is the usage bar.
- Spending: the dollars and tokens that sessions, Workers and groups have spent, with an optional budget per group.
The usage bar
The usage bar is a strip along the foot of the window. It has one block per harness that is installed on the environment's host, or that has been read before: Claude Code, Codex and opencode. Each block shows a meter of its main window and, for every window, the share used and the time left until it resets, such as 42% 2h 15m.
Hover a block for the details: the harness and its plan, a limit alert if the harness sent one ("Limit reached" or "Close to the limit"), and one line per window with its share, "limited" when the window is blocked, and when it resets.
Windows are listed in this order: 5 hours, rolling, other fixed spans, weekly, monthly, and windows the harness names itself. Windows scoped to one model come after the general ones.
The bar reads each harness's usage every five minutes while it is shown, and the refresh button, "Refresh usage", reads it now.
| Harness | Where the reading comes from |
|---|---|
| Claude Code | A short-lived Claude Code process asked for its usage, with hooks off and no session saved, and the status line of running Claude Code chats. Reads are kept at least five minutes apart. |
| Codex | Codex's app server and its rate limits |
| opencode | The opencode server; the block waits for the server to start |
| Gateway and ACP agents | Not read |
When usage is not available
Instead of a reading, a block may show a status, with a hint on hover:
| Status | Meaning |
|---|---|
| Reading… | The first reading is under way |
| Starting… | Waiting for the harness's server to start |
| Signed out | Sign in to the harness to see its subscription usage |
| No plan | The account has no subscription with usage limits, such as an API key |
| Rate limited | The harness refused to report usage for now; the app asks again later |
| Not available | The harness did not answer, or answered in a shape this version of the app cannot read; on some hosts usage cannot be read yet |
After a failed read, the block keeps showing the last good reading, greyed, with "Showing the last reading." The app waits longer between attempts after each failure, up to an hour, and respects the time a rate limit names.
Warning colour
A window turns to the warning colour once it reaches the share set in Warn at, and to the limit colour when it reaches 100% or the harness marks it as limited.
Settings
Settings › General has a card for the usage bar:
| Setting | Values |
|---|---|
| Usage bar | Shown or Hidden |
| Warn at | 50%, 75%, 80% or 90%; 80% by default |
The command palette's Toggle usage bar shows or hides it too.
Spending
Each chat's usage reports are added up per session, per Worker and per group:
- Cost comes from the harness when it reports one, as Claude Code does. Otherwise the app prices the tokens with the model's price per million tokens, from a list of known models it ships.
- Tokens count input, output, cache reads and cache writes.
Where it shows:
| Place | What |
|---|---|
| A session's row in the sidebar | The session's cost and tokens, such as $1.24 · 310k tokens |
| A Worker card in the lead's chat | The Worker's cost and tokens |
| A group in the sidebar | The group's total, with its budget when one is set, such as $12.40 / $20.00 |
Budget per group
Settings › General has a card for spending:
| Setting | Values |
|---|---|
| Budget per group | Off, $5, $20, $50, $100 or $250 |
| Warn at | 50%, 75%, 80% or 90%; 80% by default |
The budget applies to each group on its own. A notice comes once when a group's spend reaches the warning share, "api has spent $16.00 of its $20.00 budget", and once when it goes over, "api is over its $20.00 budget: $21.00 spent". The group's total in the sidebar takes the warning colour as well.
When you start a variant run, the draft warns if the run, estimated from the average cost of earlier variants of the group, would take the group near or over its budget.
Spending is an estimate from what the harnesses report. It is not a bill: your provider's own records are the reference.