Set budgets and get alerted

The key table opens. The Attribution column shows — for a key with no defaults.

The Attribution — {key name} dialog opens on the Cost attribution defaults (optional) block. Its helper reads Applied to requests on this key that carry no metadata or X-Dex-* tag of their own.

For example support, helpdesk, production. Values are trimmed and capped at 120 characters. There is no field for agent or end user here.

The row now shows chips team=support, application=helpdesk, environment=production. On failure the dialog shows the server message or save failed.

With no budgets the table reads No budgets yet — requests are not gated. Add one above. The basis line under the heading reads Basis: model spend (list price, the figure the gate reserves against). Platform overhead — classifier and embedding calls — is reported separately on the Overview and is not counted here.

The Value field is disabled and reads whole workspace.

Leave Soft threshold (%) at 80 and At 100 % on warn. Add budget stays disabled until an amount is entered.

The button reads Saving…, the form resets, and the row shows Workspace / whole workspace, $1,000.00, 80%, badge warn, and — no spend recorded this period until traffic arrives.

The field's placeholder reads exact attribution value: it must match the team tag character for character.

The Block requests at 100 %? dialog opens. Read the consequence, then click Block at 100 %. A red block badge appears next to Add budget.

The row shows Team / support, $250.00, 80%, badge block, and a switch to warn link.

With no channels the Alert channels card reads No channels — anomalies are recorded but nothing is notified.

Paste your public HTTPS endpoint into URL. Leave Secret (webhook, optional — generated if blank) empty, or type at least 16 characters. Add channel stays disabled until name, target, and a valid secret are in place.

A panel reads Signing secret — shown once. Store it with the receiver now; it cannot be read back. Click Copy, store the value with your receiver, then click Done.

The button reads Sending…, then a line under Status reads test ok with the HTTP code and latency, or test failedwith the error. The row's badge is enabled.

The hint reads spend + in-flight reservations against the budget for the current month.

Only enabled budgets are listed. With none, the select reads no enabled budgets — create one on the Budgets tab.

The input carries a % suffix. Set Re-check window to 6 hours; its hint reads utilization is over the month; the window only spaces repeat firings. Set Cooldown to 24 hours.

Name, threshold, and value reset; metric, scope, window, cooldown, and channels are kept for the next rule.

The hint reads list-price model spend over the window: the same figure the budget gate uses.

Any other scope needs a Value, and it must match one exact value.

The threshold input carries a $ prefix for this metric.

The row's Condition cell reads Spend > $25.00 over 1 hour with cooldown 1 hour under it.

The button reads Checking…, then the status line reads would fire or would not fire followed by the observed value and the window. A test never fires and records nothing.

They read the budget, Spend this period, Remaining (red and negative when over) and Forecast end of period, which reads Above the budget when the projection passes the limit. Until enough of the period has passed it reads — with the reason, for example Available after 2 days of the period, or a quarter of it if shorter. Without an organization budget the budget tile reads No limit and No organization budget.

The meter reads the share used, for example 62% used, with the spend and any in-flight reservations against the limit, then the token and request limits.

The chart draws Spend per day against the Even-burn targetfor each UTC day of the budget's current period. Without an organization budget it reads Set an organization budget to see the pace.