Guide // MandateGuard
Spend envelope vs standing policy: which should you use?
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 standing policy governs an agent's whole life with caps, an allowlist and an approval threshold; a spend envelope governs one task with its own maximum, vendors, period and written conditions, judged on a reviewer card. Use the policy for continuous operations and envelopes for work with its own budget line, client, deadline or exceptional vendors — especially whenever someone will later ask what that specific work was allowed to cost. The decisive difference is lifetime: policy changes persist and ratchet, while envelope authority expires with the task, keeping the standing mandate clean.
What is the difference between a policy and an envelope?
A standing policy is the persistent MandateGuard policy on an agent: caps, allowlist and threshold governing all its ordinary authorize calls until changed. A spend envelope is a separate, task-scoped mandate with its own maximum, vendors, period and conditions, evaluated on the reviewer card. The policy answers what may this agent ever do unsupervised; the envelope answers what may this task cost, under what terms, before it ends.
Scope: a life versus a job
The standing policy's scope is an agent's whole working life — every ordinary purchase, every ordinary day, until a human changes it. The envelope's scope is one job, bounded in money, counterparties and time. That single difference generates all the others. Policies must be conservative enough to run unattended for months, which makes them wrong for exceptional work. Envelopes can be generous inside their bounds — a large maximum, an unusual vendor — because the bounds themselves expire. When you feel pressure to make the standing policy fit an exception, that pressure is the signal: the exception wants its own envelope, and the policy wants to be left alone.
Lifetime and the permission ratchet
Policies persist, so every loosening persists — the ratchet that quietly converts this once into from now on. Envelopes end with their period, so exceptional authority self-cleans. This is why the choice matters beyond convenience: it decides what your mandate looks like in a year. Fleets governed by policies alone accumulate archaeology — allowlist entries and cap levels whose reasons retired with the projects that justified them. Fleets that route exceptions through envelopes keep a policy a new operator can read in one sitting, plus a shelf of closed envelopes explaining every exceptional grant in the fleet's history. Auditors, incident reviewers and successors all prefer the second picture.
Supervision: thresholds versus conditions
Policy supervision is numeric and standing: the approve-over threshold routes amounts to the inbox. Envelope supervision is written and task-shaped: approval conditions plus a per-spend approval level, judged on the reviewer card with the envelope's full context — spent position, period, vendor status — attached. Conditions let a mandate express judgements a threshold cannot (this vendor only for the migration week; singles over the level need the project lead). Use thresholds for the supervision your agent needs every day; use conditions when the supervision itself is part of the task's terms — client requirements, grant conditions, a director's standing instruction for one production.
Audit shape
A policy produces a continuous ledger: ideal for trend questions (how is this agent's spending evolving?) and control questions (is the mandate holding?). An envelope produces a case file: ideal for attribution questions (what was this project allowed to cost, what did it cost, who approved the exceptions?). Organisations that bill clients or report to funders feel the difference immediately — the envelope is the artefact they already wish existed when a client queries an invoice. Organisations running internal operations feel it at incident time, when the continuous ledger is the only honest narrator. Neither audit shape substitutes for the other; the instruments are complements wearing similar clothes.
A decision rule you can apply in seconds
Ask three questions of the work. Does it have its own budget line, client or deadline? Does it need vendors, amounts or conditions the standing policy should not permanently carry? Will someone later ask what this specific work was allowed to cost? Any yes points to an envelope. All no — it is the agent's ordinary life — points to the policy, with its caps and threshold tuned from receipts. The mixed case is common and fine: the agent works the task under an envelope while its ordinary duties continue under policy. What the rule forbids is the fourth option, editing the policy temporarily — there is no temporarily in a system that only changes when a human remembers to change it back.
Running both without confusion
Concurrency stays clean with two habits. First, attribution: task requests always reference their envelope ID, ordinary requests never do, so receipts sort themselves. Second, reviews: the monthly policy review asks only about standing drift (denials, approval reflexes, dormant vendors), while envelope retrospectives happen at task close. The failure mode to watch is the envelope that quietly becomes permanent — renewed without re-justification until it is a policy in exile, exempt from policy review. Counter it with a rule: an envelope may be succeeded once for a continuing task; the second continuation triggers the promote-to-policy conversation, held deliberately and recorded.
Five actions to take now
- Default to the standing policy for continuous agent operations
- Use an envelope for any work with its own budget line, client or deadline
- Never widen the standing policy to fit a temporary job
- Promote a vendor to the standing allowlist only after envelope history justifies it
- Review both surfaces monthly: policy for drift, envelopes for closure
Worked example: one agent, both instruments, no contamination
A studio's editing agent lives under a standing policy: cap 60 dollars, daily cap 150 dollars, allowlist of two stock libraries, approve-over 30 dollars — sized for its endless ordinary work. A documentary project then needs a 900-dollar archive-footage budget from a specialist archive the policy does not list, inside five weeks, with the director approving singles over 200 dollars. That is an envelope: 900-dollar maximum, the archive as its vendor, the five-week period, the 200-dollar approval level.
Both instruments run concurrently without contaminating each other. The ordinary work never notices the project; the project never inflates the ordinary mandate; and when the documentary wraps, the archive's authority ends with the envelope while the policy still describes Tuesday's stock purchases exactly as before.
Frequently asked questions
Can a request be judged by both?
Standing requests go through the policy's authorize flow; task requests reference their envelope for evaluation. The design intent is one governing instrument per spend, with all decisions in one receipt chain.
Which is cheaper to run?
Both ride the same workspace. The real cost difference is administrative: policies are set once and reviewed monthly; envelopes are defined per task in minutes.
What if a task becomes permanent?
Promote its shape into the standing policy deliberately — that is what policy reviews are for — and close the envelope.
Which do auditors prefer?
They prefer the match between instrument and work: standing stories for operations, envelopes as self-contained case files for projects.
Can envelopes exist on the free plan's single policy?
Envelopes are a separate MandateGuard surface (/envelopes). The free plan's core is one agent and one policy; check your plan for envelope availability.
Keep reading
- What is a spend envelope for an AI agent?
- Why use a task-specific mandate? One task, one envelope
- What should an AI agent spending policy template include?
Put the policy in force: spend envelopes · intent registry · approval inbox · how the proof works · all MandateGuard guides