Spending
Model calls are priced on this device, as they happen, from the provider's own figure. The bill arrives in front of you as it is spent, wherever the provider reports a figure, and not later as a surprise.
Two pots, kept apart
The two pots are different accounts in different currencies, so they are shown as two tracks and not added together.
- Inference covers the model calls billed to your key. They are recorded here per turn, and none of them passes through us.
- Credits holds the prepaid balance, spent on the model when Daimond runs it for you and on what leaves the browser, web, mail, sync, and cloud storage.
The foot of the rail carries the running figures for today, this week and this month. Press one to open the Spending panel, which shows a headline figure per pot, a daily graph, a breakdown by category, and a plain table of movements.
What stops a runaway
A workspace that runs many agents at once can spend quickly, and the danger lies in a sudden rate more than in a large monthly total. Four measures catch the rate, and none asks you to set a limit in advance.
- A pace check before a fan-out. The cost of dispatching a batch is known before any of it starts. A batch that would run faster than your own recent pace is held, and Daimond asks once, with the number on the prompt. A few helpers of ordinary cost do not trip it, and a sudden fifty do.
- Pause and stop, always to hand. The Agents panel pauses everything running, or stops it. The Everything row at the top of the rail pauses the whole app. A paused thing refuses a turn before it starts, so a look costs nothing.
- A separate key for every agent. Each helper running on credits spends its own freshly minted key, capped against what is left after its siblings' claims. No two share a key, so parallel helpers cannot race one allowance past its cap.
- Everything metered, per turn, on your device. Every call is priced and recorded as it happens, and the graphs are drawn from that record.
A ceiling you set
Settings carries a per-turn spend cap. A turn that reaches the round limit continues on its own, so this cap is the ceiling that holds a runaway. It applies where the provider reports what a round cost, and only there.
Watching one agent
Each tile in Agents shows its cost as it accrues, in tokens and in money, and not only at the end. Under the header a line says how many are running, paused and queued, which makes the size of a fan-out legible before you decide whether to hold it.
The aim is a team of agents you can run without fear that a loop, or a fan-out larger than you realised, can cost more than a glance and a click can undo.