Spend commands

anycloud cost

anycloud cost [workload-id] [options]

Option	Description
-p, --period PERIOD	Whole days from 1d to 90d; default 30d
--session ID	Session ID, prefix, label, or agent:hash
--agent NAME	One detected agent
--user USER	One GitHub username or account ID
--json	Compact cost report

Without an ID, the report includes Jobs, Services, and VMs. A long-running Service or VM overlapping the window is included even if it started earlier, with duration and cost clipped to the overlap. Aggregate JSON includes totalJobs, totalServices, totalVms, and Job / Service / VM counts in each byProvider, byAgent, and byUser entry.

On an API shared with other GitHub users, the report covers the whole API and adds a per-user table, so a shared total is not mistaken for one person's spend. --user narrows to one account and accepts a username or an account ID; the ID also matches work submitted before a username change, which the username does not, because it is captured once at submit time. The human-readable per-user table still appears when a teammate is the window's only contributor and is omitted only when all reported spend belongs to the caller.

Cost follows provisioned VM lifetimes: accrual starts when a VM is created and stops at its endedAt cutoff. A workload's completedAt is its outcome timestamp and neither selects nor caps cost reporting.

Billing reconciliation uses the compute credential recorded on each VM attempt. Malformed unrelated SSH, storage, container payloads, or requested placement settings no longer prevent available provider costs from settling. Missing required workload or container-spec structure and unusable compute credentials still leave the affected costs unsettled; other billing groups continue.

Billing processes historical work in small batches while keeping each resource's credential group together. One billing query covers every group in a batch that shares a credential, and each group's costs commit as one unit; an interruption leaves a group fully saved or ready to retry, preserving its cost allocation. A batch the provider rejects as malformed is retried one group at a time. One unusually large group can still delay requests. A group stops being queried a week after its last VM ended or its owner finished or was deleted, whichever is later. A credential the billing API rejects with 429 waits until the next hourly run. VMs left unbilled keep showing their estimated cost.

Window reports are aggregated in SQLite and are not capped by workload count. Their response size depends on provider, agent, and user cardinality rather than the number of matching workloads; session, agent, and user filters scope the accounting rather than prevent a partial or rejected total.

anycloud throttle

anycloud throttle set [usd-per-hour] [--agent-session]
anycloud throttle show
anycloud throttle unset [--agent-session]

Option	Description
--agent-session	Set or remove an independent cap for each detected session

Throttle gates a candidate when estimated running VM dollars per hour plus the candidate is greater than or equal to the cap. Running workloads continue.

anycloud budget

anycloud budget set [usd] --per <day|week|month> [--agent-session]
anycloud budget show
anycloud budget unset --per <day|week|month> [--agent-session]

Option	Description
--per PERIOD	Required UTC window: day, week, or month
--agent-session	Set or remove a per-session budget

anycloud spend

anycloud spend show [--json]

--json adds percentUsed and headroomUsd fields and omits terminal bars and color.

On an API shared with other GitHub users, human-readable anycloud spend show prefixes each agent-session row with the account that ran it, so a session consuming the shared cap says whose run it is. The prefix is omitted when every session in that cap section belongs to the same account. Work submitted without an agent session has no row here; anycloud cost reports that spend per user.

See Spend Controls for dispatch behavior and scope.