Guide // MandateGuard
What is a mandate receipt? Hash-chained proof for agent spend
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 mandate receipt is the signed record of one agent spend decision: vendor, amount, task reference, the decision ALLOW, DENY or REQUIRE_APPROVAL, and the reason. In MandateGuard, receipts form a hash chain — each receipt's hash is computed from the previous receipt's hash and the receipt content, then HMAC-signed — so editing any stored row breaks verification at that sequence. The verify-chain endpoint recomputes the ledger on demand. Receipts turn agent spending from something you reconstruct after an incident into something you can prove, decision by decision, including the spends that were refused.
What is a mandate receipt?
A mandate receipt is a ledger row recording an agent spend decision: vendor, amount, task reference, decision and reason, plus a sequence number. In MandateGuard each receipt's hash is computed from the previous receipt's hash together with the canonical receipt content, and the row is HMAC-signed server-side. Verification recomputes the whole chain from stored rows; tampering with any stored field breaks verification at that sequence number.
Why a chain, and not just a log
Ordinary logs can be edited quietly: a row deleted, an amount adjusted, a denial removed before a review. For spending authority, that weakness matters, because the log is the thing you will rely on exactly when someone has an incentive to change it. Hash chaining removes quiet edits. Each MandateGuard receipt commits to its predecessor, so changing receipt 41 changes its hash, which breaks receipt 42's commitment, and so on to the head of the chain. The verify endpoint recomputes from stored rows and tells you whether the ledger you are reading is the ledger that happened. It is the same structural idea as a blockchain's linking, applied privately to one workspace's decisions.
What each receipt records
A receipt is deliberately boring: a sequence number, the vendor, the amount in cents, the task reference the agent supplied, the decision, the reason, the hash of the previous receipt and its own hash, signed server-side. Boring is a feature. An auditor does not need the agent's chain of thought; they need to know what was proposed, what the mandate answered, and whether the record can be trusted. The task reference is the joining key between the receipt and your own systems — tickets, jobs, campaigns — which is why agents should supply meaningful references rather than placeholder strings. Good references make the difference between a ledger that explains itself and one that needs archaeology.
Denials are evidence, not embarrassment
Teams new to receipted spending sometimes read denials as failures. They are the opposite: a denial receipt is proof the mandate intercepted an overspend, a wrong vendor or a frozen agent before money moved, with the attempted amount preserved. In an incident review, the denied receipts often carry the story — the retry storm visible as repeated identical denials, the lookalike vendor attempt visible once, the oversized subscription visible with its full annual price. Deleting or ignoring them would leave only the spends, and a record containing only spends cannot distinguish a well-run agent from an unobserved one.
Verification in practice
MandateGuard exposes chain verification as an endpoint that recomputes every hash from the stored rows and reports a verdict plus the receipt count. Run it on a rhythm — weekly is reasonable — and always after an incident, a migration or a dispute. A verified verdict means the chain is internally consistent end to end. A broken verdict names where verification fails, which focuses investigation on a specific sequence rather than the whole ledger. Paid plans can export an evidence pack: the receipts plus the chain verification, packaged for auditors and insurers who were not present when the decisions were made and should not have to take your word for them.
Receipts beyond MandateGuard
Receipts also travel. The interoperable intent service lets a merchant-role key verify a specific receipt against a registered intent — checking hash, signature, workspace chain and intent linkage — without holding any write access. That cross-party verification is what makes a receipt more than an internal log line: the merchant, the agent's operator and an issuer-side checker can all confirm they are looking at the same decision. Standalone receipt export for arbitrary agent stacks is a separate product in the fleet (Agentic Receipts); MandateGuard receipts are native to its own authorize flow and intent events.
Handling, retention and privacy
Receipts contain operational data — vendors, amounts, task references — not payment credentials, because MandateGuard never moves money. Store exports where you store other financial evidence, keep them for the period your auditors require, and be thoughtful about task references: they are visible to anyone you share the pack with, so reference ticket numbers rather than pasting customer details into them. Workspace data can be exported or deleted on request to the support address listed in the privacy policy, and license keys and tokens throughout the system are stored only as hashes, following the same minimise-what-you-hold discipline.
Five actions to take now
- Create the workspace, agent and policy, then run the demo to see real receipts
- Record receipt sequence and hash in your own task logs
- Call verify-chain weekly and after any incident
- Export the evidence pack before auditors ask, so you know the format
- Treat denied receipts as records too — they prove the mandate worked
Worked example: 212 receipts settle a dispute
Over one quarter a support agent accumulates 212 receipts: 184 allows, 19 approvals granted, 9 denials. When its workspace is reviewed, the chain verifies intact at 212 entries — every amount, vendor and decision exactly as recorded at the time. One denial in February shows a 480-dollar attempt from an allowlisted vendor against a 100-dollar per-task cap.
That single receipt settles an argument. The vendor insists the agent ordered the enterprise tier; the receipt proves the engine refused it within the mandate, before money moved, and the allowed receipts show the monthly plan the agent actually bought at 40 dollars. The review takes an afternoon of reading, not a month of reconstruction, because the evidence was created at decision time.
Frequently asked questions
What fields does a receipt hold?
Sequence, vendor, amount, task reference, decision, reason, hash, previous hash and signature — enough to recompute and audit the entry independently.
Are denied attempts really receipted?
Yes. A denial is a decision. Its receipt records the attempted amount and the bound that stopped it.
Can I verify the chain myself?
The verify-chain endpoint recomputes the chain from stored rows and reports its verdict and count. Paid plans add evidence export packaging receipts with that verdict.
What breaks verification?
Editing any stored amount, vendor or decision changes the recomputed hash at that sequence. The chain is tamper-evident by construction.
Do receipts contain payment credentials?
No. MandateGuard decides on spends; it never moves money and stores no agent payment credentials. Receipts are decision records.
Keep reading
- How do you verify a hash-chained receipt ledger?
- How do you audit AI agent spending?
- What is an evidence export for agent spending audits?
Put the policy in force: spend envelopes · intent registry · approval inbox · how the proof works · all MandateGuard guides