Skip to main content

Usage & billing

The Usage page shows what your organization is spending on agent activity — model requests, agent runs, tool calls, and related surfaces — and lets admins set budget policies to cap internal spend. Customer-visible contracts define entitlement; org budgets define house rules inside that entitlement.

Reading usage

Usage is shown over a selectable time window (7 / 30 / 90 days or custom) with rollups by org, team, project, member, model, and repository where your plan exposes attribution.

Included vs on-demand

Two buckets appear on every org usage summary:

BucketMeaning
IncludedSpend covered by your plan's monthly allowance from contract / capacity
On-demandSpend beyond the included allowance — billed or throttled per your commercial terms

The included allowance (monthlyUsdCap from entitlements) is set by platform operators in the Console. Admins see the same cap on Usage and in Settings budget dialogs as the reference line for "plan included."

What drives spend

Event kindTypical attribution
Model inferenceTokens × model cost class
Agent runsRun duration, tool calls, and model usage combined
Cloud agent computeIsolated environment time plus model usage
Policy denialsNo spend — but logged for troubleshooting

Use filters for surface (desktop, CLI, cloud), project, and model when investigating spikes after a release or CI change.

Usage reservations at run time

Before a cloud agent run starts, agents-api.4rged.ai calls reserveAgentRunUsage to preflight spend against org budget and contract entitlements. If the reservation is denied, the API returns 402 Payment Required with a usageReservation payload explaining the block — no job is enqueued. Successful reservations record estimated_usd and reconcile to reconciled_usd when the run completes; Sentinel and the operations summary aggregate these rows for the last 24 hours.

Budget policies

Budgets are your internal caps — stricter than contract, never looser. Enforcement modes map to how the platform responds as you approach or exceed a cap:

EnforcementBehavior
ObserveTrack against cap; no member-facing block (reporting only)
WarnNotify admins/members; runs generally continue
Soft stopBlock or queue new on-demand spend; may allow included bucket until depleted
Hard stopBlock new agent runs that would incur spend until cap resets or admin raises policy

Org settings UI often expresses on-demand behavior as unlimited (warn), fixed cap (soft stop), or strict cap (hard stop) — saved server-side as monthlyUsdCap plus enforcement enum.

Setting a budget policy

  1. Open Manage budget

    From Usage (budget meter card) or Settings → Organization, open Manage budget. Requires admin or super admin.

  2. Choose cap mode

    ModeUse when
    Monthly USD capStandard org-wide house rule aligned to calendar month
    Range capShort experiments or release windows with a fixed total

    Enter a USD value at or below your contract ceiling. Setting above entitlement may clamp to contract maximum in UI.

  3. Set enforcement

    Pick how aggressively to enforce as usage approaches the cap:

    • Warn for learning phases and internal dogfood orgs
    • Soft stop when you want to stop on-demand burn but avoid hard outages mid-incident
    • Hard stop for strict cost control in production orgs
  4. Save and communicate

    Save the policy — it applies organization-wide and is audited. Notify team leads when moving from warn → hard stop so CI and cloud automations are expected to fail closed.

Team and project budgets

Enterprise layouts may allocate subsets of org budget:

ScopePurpose
Org budgetTop-level cap and default enforcement
Team budgetSquad allocation — platform vs product spend
Project budgetInitiative cap — e.g. migration project burn limit
Member limitsTargeted warnings for heavy individual usage

Team/project budgets narrow org capacity — they cannot grant spend beyond org policy or contract.

Admin walkthrough: monthly review

  1. Compare to included allowance

    On Usage, confirm Included consumption vs plan allowance. If included is consistently exhausted mid-month, request capacity from your operator via Console — do not only raise internal budget without entitlement.

  2. Inspect on-demand drivers

    Sort attribution by project and model. Common findings: CI using a reasoning model, cloud migration project, or new team without team-level cap.

  3. Review denials

    Spike in denials may indicate model policy misconfiguration rather than budget — cross-check Model control.

  4. Adjust policy or catalog

    Tighten model groups, lower project budgets, or switch CLI default to a lower cost class before raising org cap.

  5. Export

    Export CSV for finance or chargeback when your plan includes usage exports.

Budget vs contract — quick reference

ConceptSet byGoverns
Contract / capacityOperator (Console)What the org is entitled to
Included allowanceOperator (Console)Size of the Included bucket
Budget policyOrg admin (Dashboard)Internal USD cap and enforcement
On-demand billingCommercial termsSpend beyond included
Budgets vs. contracts

Budgets are your internal house rules (set by admins). Contracts and capacity are your entitlement (set by operators in the Console). Budgets can be stricter than your contract, never looser.

Hard stop in CI

Hard stop blocks new runs that would incur spend — scheduled CI jobs fail until the cap resets or an admin raises policy. Prefer warn or soft stop during pipeline bring-up.