Guides // MandateGuard
MandateGuard guides: agent spending controls, explained
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.
Every guide below answers one real question about giving AI agents a budget without giving them the company card: mandates, caps, allowlists, approvals, receipts, intents and envelopes — grounded in what MandateGuard actually does.
What is an AI agent spending mandate?
A spending mandate is a signed, scoped permission that decides what an AI agent may buy, from which vendors, and when a human must approve. Here is how mandates work in MandateGuard.
How do you set spending limits for an AI agent?
Set agent spending limits that survive retries and edge cases: per-task caps, daily caps, vendor allowlists and approval thresholds, sized from real prices and enforced before payment.
What is a per-transaction spending cap for an AI agent?
A per-transaction (per-task) cap is the largest single amount an agent may spend at once. How it works, how to size it, and how it interacts with daily caps and approvals.
How do daily spending caps stop AI agents leaking money?
Daily caps bound an agent's total spend per day, catching retries, loops and slow leaks that per-transaction caps miss. How MandateGuard counts the day and how to size the cap.
How does a vendor allowlist work for AI agent spending?
A vendor allowlist names the only sellers an agent may pay. How MandateGuard enforces it, why it beats blocklists, and how to keep the list tight without blocking real work.
When should a human approve an AI agent's spending?
Approval thresholds route the right agent purchases to a human: big enough to matter, in-policy enough to consider. How REQUIRE_APPROVAL, the inbox and one-time tokens work.
ALLOW, DENY or REQUIRE_APPROVAL: how agent spend decisions work
Every MandateGuard authorize call ends in one of three decisions. What each means, which check produces it, what the agent should do next, and how each is receipted.
What is a mandate receipt? Hash-chained proof for agent spend
A mandate receipt records every agent spend decision in a hash-chained ledger. How chaining works, what verify-chain proves, and why denials belong in the record.
How do you stop an AI agent from overspending?
Agent overspending has five shapes: one big purchase, many small ones, the wrong vendor, the duplicate, and the loop. The control for each shape, and how MandateGuard enforces it.
How does an AI agent kill-switch work? Freezing spend instantly
The agent kill-switch (freeze) refuses every spend instantly while preserving the full receipt history. When to pull it, what happens to in-flight work, and how to unfreeze safely.
How does duplicate-commitment detection stop agents paying twice?
Duplicate-commitment detection flags an agent buying the same vendor-and-amount commitment twice inside a window. Why agents duplicate, and how the ledger catches it.
What is vendor trust scoring from your own receipt history?
MandateGuard vendor trust is computed only from your own receipt history: allowed and denied counts and totals beside every score. Why first-party-only data is the point.
How do you add spending controls to a purchasing agent?
Adding spending controls to a purchasing agent: its own policy, a tight allowlist, approvals for first orders, task references that join receipts to purchase orders, and a weekly rhythm.
How do you require human approval before an AI agent pays?
Requiring human approval before an agent pays: the approve-over threshold, the inbox decision, and the bound one-time token that makes approval unforgeable and unreusable.
How do you integrate an authorize check before an agent spends?
Integrating the authorize API: workspace, agent and policy setup, the request the agent sends, the three decisions, approval redemption, and the integration mistakes that void the whole control.
How do MCP tools let an agent check its spending mandate?
MCP tools let agents check spending policy natively: mandate.preview before a spend, mandate.get_receipt for proof, plus the intent tool family. Same engine, same ledger.
How do you audit AI agent spending?
Auditing AI agent spending from signed receipts: reconciling the hash-chained ledger to invoices, reading denials as control evidence, and preparing the pack an auditor actually wants.
What is an evidence export for agent spending audits?
An evidence export packages agent spend receipts with the chain verification for auditors and insurers. What belongs in the pack, when to produce it, and the questions it must answer.
How do webhook notifications work for agent spend approvals?
Approval webhooks notify your systems when an agent spend waits for a decision. Where webhooks fit, what they must never do, and how to wire them into a daily ops rhythm.
How does a one-time spend token work after approval?
A one-time spend token turns a human approval into exactly one purchase: bound to the approved vendor and amount, stored hashed, redeemable once. Why binding matters.
How do you freeze a misbehaving AI agent?
Freezing a misbehaving agent: the signals that justify it, the exact sequence from freeze to unfreeze, and the mistakes that turn a contained incident into a recurring one.
How do you set up your first AI agent spending policy?
Setting up your first agent spending policy in five steps: workspace, agent, prices, the four settings, and the go-live checks that prove the mandate actually governs the agent.
How do you verify a hash-chained receipt ledger?
Verifying a hash-chained receipt ledger: what the recomputation proves, what a broken verdict means, when to run it, and how verification anchors audits and disputes.
What should an AI agent spending policy template include?
A complete agent spending policy template: the four enforceable settings, approval conditions, review rhythm and worked numeric examples for research, purchasing and provisioning agents.
What is an AP2-style intent mandate for agent payments?
An AP2-style intent is a registered, referenceable mandate: principal, agent, merchant scope and budgets under one intent ID that agents, merchants and issuers can all check.
What is an interoperable intent service for agent mandates?
The interoperable intent service explained: one registered intent state that agent, merchant and issuer-side checkers read independently, with role-scoped keys and receipt verification.
How do cumulative budgets work on an agent intent?
Cumulative budgets cap the total an intent can ever spend. How atomic decrement, refund events and the budget gauge keep a shared total truthful across parties.
How do per-period budgets and authority windows pace an intent?
Per-period budgets pace an intent's spending (a cap per 30 days); authority windows bound its life. How the two compose with cumulative budgets on a registered intent.
How do role-scoped keys work for issuers, merchants and agents?
Role-scoped keys give each party to an intent exactly its rights: issuers write, merchants and agents read and verify. Minting, labels, revocation and the mistakes to avoid.
How do merchants verify an agent's receipt against an intent?
Merchants verify agent receipts against an intent in four checks — hash, signature, workspace chain, intent linkage — using a read-only key. When to check and what failures mean.
How do refunds, adjustments and chargeback events work on an intent?
Intent events record post-transaction reality — refunds, adjustments and chargeback notes — in the intent's own hash chain, restoring or correcting its budget for all parties.
When should you suspend, resume or revoke an agent intent?
Managing an intent's life: suspend to pause reversibly, resume to restore, revoke to end. When each is right, what checkers see, and the communication that keeps it clean.
What is the EMVCo draft framework for agentic payments?
The EMVCo draft framework for agentic payments, explained in plain terms: why payments bodies are standardising agent mandates, and what MandateGuard's draft alignment does and does not claim.
AP2, UCP and x402 vs a spending-policy layer: what is the difference?
AP2, UCP and x402 prove a payment was authorised; a policy layer decides whether it should be. How the two compose, and where MandateGuard sits in the stack.
What is a spend envelope for an AI agent?
A spend envelope is a mandate for one task: maximum total, approved vendors, period and approval conditions, judged on one reviewer card. When envelopes beat standing policies.
Why use a task-specific mandate? One task, one envelope
Task-specific mandates scope spending authority to one job: its own total, vendors, period and conditions. Why scoping by task beats stretching a standing policy.
How does the reviewer card decide within, needs approval, or outside mandate?
The reviewer card puts one spend request against one envelope on a single screen: within mandate, needs approval, or outside mandate. How to read each state and decide well.
How do you define a spend envelope, step by step?
Defining a spend envelope step by step: the task sentence, maximum total, vendor list, period and approval level — plus the test evaluation that proves the mandate reads as intended.
Spend envelope vs standing policy: which should you use?
Standing policy vs spend envelope: scope, lifetime, supervision and audit compared, with decision rules for which instrument fits continuous ops, projects and client work.
How do you write approval conditions on a spending mandate?
Approval conditions are the written rules that decide when a spend needs a human and what the human checks. Writing conditions that evaluate cleanly, with examples that work.
How do you control spending for AI agents that buy cloud services?
Controls for agents that buy cloud services: caps fitted to stack prices, approvals for full environments, and reconciling mandate receipts to the cloud bill to find ungated paths.
How do you control AI agent subscription spending?
Controlling agent-started subscriptions: approvals on starts, duplicate detection on repeats, renewal visibility in receipts, and the zombie-subscription review that pays for it all.
How do you budget a data enrichment agent's API spending?
Budget controls for data enrichment agents: batch-sized caps, provider allowlists, duplicate protection against buying the same data twice, and cost-per-record reviews.
How do you set spending limits for a travel booking agent?
Spending limits for travel booking agents: channel allowlists, per-booking caps, approvals for changes, and the cancellation freeze that stops a dead trip from spending.
How do you set guardrails for AI agent advertising spend?
Guardrails for AI agents that spend on advertising: campaign envelopes, approval levels for budget shifts, platform allowlists and weekly reconciliation against platform reporting.
How do you give a research agent an API budget that works?
Budget guide for research agents: daily-cap-led control for many small API buys, approvals at bulk boundaries, and joining receipts to research output so unused sources get cut.
How do you manage spending policies for a team of AI agents?
Running spending policies for a team of agents: per-agent policies, coordinated daily caps, one approval discipline, and the workspace-level controls that catch cross-agent waste.
What spending controls do finance teams need for AI agents?
Agent spending controls for finance teams: delegations of authority as caps, approvals as countersignature, reconciliations that find ungated paths, and audit-ready evidence.
How do Agent Passport and spending mandates work together?
Agent Passport (identity) and MandateGuard (spending authority) compose: attach a passport ID to the agent record, and let each system answer the question it owns.
How do you set guardrails for autonomous procurement agents?
Guardrails for autonomous procurement: approved-supplier allowlists, commitment approvals, duplicate detection across categories, and reconciling agent commitments to purchase orders.
AI agent spending policy vs prompt instructions: what actually enforces?
Prompt instructions guide an agent's judgement; a spending policy enforces boundaries outside the model. Why money rules belong in the engine, and what prompts are still for.
Why don't payment rails alone limit AI agent spending?
Payment rails prove a payment was authorised; they do not decide whether it should happen. What rails prove, what they cannot see, and where the limiting layer belongs.
What should an AI agent spending audit trail include?
Requirements for an agent spending audit trail: completeness (denials included), contemporaneity, integrity by hash chain, joinability to business records, and portable evidence.
How do you prevent duplicate subscriptions from AI agents?
Preventing duplicate agent subscriptions: gate the starts, detect the repeats from the ledger, and keep a renewal tracker so the business always knows what it is committed to.
What is the playbook for a runaway AI agent spending incident?
The runaway agent spending playbook: freeze first, read the chain to the first anomaly, classify the cause (loop, tool change, steering, policy misfit), fix, restore, record.
Free vs Pro vs Team: which MandateGuard plan fits your agents?
Free vs Pro vs Team for agent spending controls, honestly compared: Free runs one agent's full control loop; paid adds multi-agent, duplicates, trust, kill-switch, export and intents.
How do you choose AI agent spending control software?
Choosing agent spending control software: the criteria that matter — external enforcement, three-state decisions, verifiable evidence, bound approvals — and the demo that reveals them.
What spending controls does a startup need for its AI agents?
Agent spending controls for startups: start with the riskiest agent on free, fit caps from real charges, one approver with a backup, and the habits that scale into a fleet.
What should a monthly AI agent spend review include?
The monthly agent spend review routine: verify the chain, read denials and approvals, check subscriptions, tune one bound per agent. An agenda that runs in under an hour.
What are the most common AI agent spending policy mistakes?
The common agent spending policy mistakes: one-cap policies, guessed numbers, bloated allowlists, dead thresholds, unstaffed inboxes — and the receipt evidence that exposes each.
Glossary: AI agent mandate and spending terms explained
A glossary of agent mandate terms as MandateGuard uses them: mandate, caps, allowlist, decisions, receipts, tokens, envelopes, intents, role keys — defined by what each does.
What are the best practices for AI agent budget governance?
AI agent budget governance best practices: grant narrowly, bound every dimension, supervise the top band, evidence everything, and re-justify delegations on a schedule.
How do you give an AI agent a budget safely?
How to give an AI agent a budget safely: turn the amount into a mandate with caps, allowlist, approvals and a freeze — and the five mistakes that make budgets unsafe.