Guide // MandateGuard
How do cumulative budgets work on an agent intent?
By the Ingenious AI team · Published 2026-11-14 · Reviewed 2026-11-14
▸ Run the free live demo See pricing
Free covers 1 agent and 1 policy with the full authorize flow, signed receipts and the approval inbox — no card required to start.
Short answer
A cumulative budget is the total an intent may spend across its whole life: every ALLOW against the intent decrements the remaining budget atomically, and refund or adjustment events restore it, so the spent and remaining figures every party reads are always current. It answers the question period budgets cannot — how much, in total, has this mandate committed? Watch the gauge's burn rate against the authority window, append return events promptly, and treat the approach to exhaustion as a scheduled renewal decision rather than a surprise at checkout.
What is a cumulative budget on an intent?
A cumulative budget is the maximum total spend an intent may accumulate across its whole life or authority window. In MandateGuard each ALLOW against the intent atomically decrements the remaining cumulative budget, and refunds or adjustments appended as events restore or correct it. The intent shows spent, remaining and percentage consumed on its gauge, identically to the issuer, the agent and any merchant checking it.
Why a lifetime total, when periods exist
Per-period budgets pace spending; cumulative budgets bound commitment. A mandate can be perfectly paced and still total more than the principal ever agreed to — period after period of legitimate spend adding up past the mandate's intent. The cumulative figure is the principal's actual promise: in total, across this whole arrangement, no more than this. It is also the number counterparties reason with. A merchant extending mandate-backed terms wants the arrangement's total exposure; an issuer's risk review wants the maximum the intent can ever draw. Period budgets answer how fast; the cumulative budget answers how much, ever — and both questions deserve explicit numbers.
Atomic decrement, explained plainly
The dangerous moment for any shared budget is simultaneity: two purchases in flight, each checked against the same remaining balance, each allowed, together overshooting. MandateGuard decrements the intent's spent total atomically as part of a successful authorisation — the check and the decrement are one operation, so the second purchase always sees the first's effect. This is the same decision-based counting the policy engine uses for daily caps, extended to shared state. It sounds like an implementation detail until the first time two of an agent's parallel workflows buy at once; then it is the difference between a budget that is a fact and a budget that is an estimate.
The gauge as a shared instrument
The intent's budget gauge — spent, remaining, percentage consumed — is shown identically in the registry to every authorised viewer. That symmetry disciplines everyone. The agent's operator can see a mandate running hot before the wall. The issuer can compare burn rate against the authority window: 80 percent consumed with 30 percent of the window elapsed is a forecast, not a surprise. Merchants can judge whether a large incoming order fits the remaining budget before promising fulfilment. Shared numbers end the most corrosive pattern in mandate relationships, where each party privately assumes the budget is healthier than it is and the discovery happens at a failed checkout.
Keeping the total honest: events
A cumulative budget is only as truthful as its maintenance. Purchases decrement it, but commerce also runs backwards: goods are returned, prices are adjusted, charges are disputed. The intent event log exists for exactly this — refund and adjustment events appended against the intent restore or correct the spent total, hash-chained with the authorisations so the arithmetic's history is auditable. The operational habit this demands is promptness: the operator who learns of a refund appends the event the same week, not at quarter end. Stale budgets mislead every party reading them, and in a shared system your staleness becomes other people's planning errors.
Exhaustion is a process, not a surprise
When cumulative spend reaches the total, authorisations under the intent cannot proceed — that is the budget doing its final job. Well-run relationships treat the approach to exhaustion as a scheduled decision point: at a agreed threshold (80 or 90 percent), issuer and operator review whether the mandate renews, expands as a new intent, or winds down. Because status changes are explicit, winding down is clean: suspend to pause while goods in flight settle, then revoke when the arrangement truly ends. The event log preserves the whole life of the intent after revocation, so the historical record remains verifiable even though the authority is gone.
Cumulative budgets among the other controls
Within a workspace, an intent's cumulative budget composes with — never replaces — the agent's policy: authorisations still pass caps, allowlist and thresholds, and still produce receipts. Across parties, it composes with per-period budgets (pacing), merchant scope (counterparties) and the authority window (time). A useful mental model is nested envelopes of permission: the policy is the agent's standing law, the intent is a contract's total, the period is its rhythm, the window is its lifespan. Setting all four consciously, from the commercial reality of the arrangement, is what makes a shared mandate feel less like a leap of faith and more like bookkeeping.
Five actions to take now
- Set the cumulative budget from the mandate's full expected cost, not a hopeful one
- Pair it with a per-period budget when pacing matters
- Watch the gauge: burn rate against the authority window is your early warning
- Append refund and adjustment events promptly so remaining budget stays honest
- When the budget nears exhaustion, renegotiate deliberately — register new terms rather than stretching old ones
Worked example: a 2,400-dollar intent across 90 days
An issuer registers a 90-day intent with a 2,400-dollar cumulative budget for a facilities agent buying supplies from two scoped merchants. Month one spends 610 dollars. A 90-dollar delivery is returned in month two: the refund event restores it, so spent shows 520 dollars carried into the remaining period plus new purchases of 480 dollars — month two closes at 1,000 dollars cumulative.
In month three a project pushes spending to 2,310 dollars. The gauge's 96 percent reading prompts a deliberate conversation a week before exhaustion: the issuer registers a follow-on intent for the next quarter rather than letting the agent hit a wall mid-order. Cumulative budgeting turned a potential failed purchase into a scheduled renewal, because the total was visible to everyone all along.
Frequently asked questions
What does atomic decrement prevent?
Concurrent authorisations each seeing the same remaining budget and jointly overshooting. The decrement happens as part of the ALLOW, serialised by the engine.
Do refunds restore cumulative budget?
Yes, when appended as refund events against the intent — which is why prompt event hygiene matters for shared mandates.
Can the cumulative budget be raised in place?
Budgets are registration terms. Changing them mid-flight undermines the shared record; register a successor intent for new terms.
How is this different from a daily cap?
A policy daily cap bounds one agent's spend per day inside your workspace. A cumulative budget bounds the whole intent's life and is visible to outside parties.
What if spend must stop before the budget ends?
Suspend the intent (reversible) or revoke it (final). Status changes take effect for every checker at once.
Keep reading
- How do per-period budgets and authority windows pace an intent?
- What is an AP2-style intent mandate for agent payments?
- How do refunds, adjustments and chargeback events work on an intent?
Put the policy in force: spend envelopes · intent registry · approval inbox · how the proof works · all MandateGuard guides