AXR/Plain English/For people who run AI, not cryptography
Your AI books appointments, sets prices, sends messages, processes refunds — real money, real decisions, real risk of a dispute. The day a customer says it did something it shouldn’t have, your log says everything’s fine. But you wrote that log. Who’s going to believe you?
AXR gives every decision a receipt — like the seal on a parcel, but for your automation. It can’t be changed afterwards, and anyone can check it without taking your word for it. Don’t take ours either: break one yourself, just below.
See it
Below is a real, signed AXR receipt — the kind your automation would produce for one booking. The seal is green because every character matches the signature. Edit any field and watch the seal break, live, on your machine. Nothing is uploaded.
AI booking receipt
Signature valid
Edit any field above. The seal breaks the instant one character no longer matches what was signed.
What just happened?
Your browser recomputed the receipt’s Ed25519 signature over the exact text shown — the same check a verifier runs. Match → green. One byte off → red. No server, no upload.
This proves the record wasn’t changed since signing. It does not prove the decision was right — it makes a wrong one undeniable.
What it actually is
When a courier hands over a parcel, they get a signature. You don’t write it; you can’t change it later; and if there’s a dispute, you both look at it. AXR is that, done automatically for your AI — and the stamp can’t be swapped out afterwards.
Written by you, on your server, at any time. In a dispute it’s your word about your own systems.
Sealed when the decision happened, checkable offline by anyone — no access to you or your systems needed.
A log you alone can rewrite is a story. A receipt anyone can check is a record.
Under the hood it’s standard, proven cryptography — a digital signature and a tamper-evident chain, nothing experimental. You never have to see any of it to use it. (If you want to check, the code is open source.)
Is this you?
Nobody searches for "tamper-evident audit trail." They hit a moment like one of these — and suddenly the question is can you prove it?
"I returned the parcel two weeks ago — your system says the refund went out. It didn’t."
"Your system dunned an invoice I already paid. Three times. Fix it or I call a lawyer."
"Your AI rejected me because I’m 52." Now prove age was never an input.
"Show that last year’s automated decision followed the policy in force at the time."
"The contract says something else than we agreed." — "But the AI wrote it…" is not a defence.
Four stories
Your automation says refund_processed: true. The customer never got the money; the bank says it never received a request. Whose fault is it? With a receipt for that step, you can show whether the bot actually issued the instruction — so you know in seconds where the chain broke, instead of eating a chargeback to keep the peace.
An AI kept sending payment demands on a settled invoice, and the customer is threatening to sue. Your internal log says everything was correct — but you wrote that log. A signed receipt for each billing decision is something the other side (or a court) can check without phoning your developer.
A rejected applicant suspects the AI screened them out for a forbidden reason. You need to show what the decision actually saw. A signed decision record lets an authority check that the banned data was never an input — they don’t have to take your word for it. (And personal data can still be erased later without breaking the signature.)
A supervisor asks you to prove a past automated decision followed the rules in force at the time. The model has changed since. An anchored, signed trail lets them verify the record was not edited after the fact — and you can hand them an auditor-ready report instead of raw files.
Not a thought experiment: AXR runs in production today on a live booking workflow — 200+ signed receipts, anchored hourly since June 2026. See the technical account.
The moment it matters
You hold up a screenshot of your log. The judge asks: "You wrote this yourself, correct?" With a receipt, the answer changes — because the other side can check it without trusting you. These moments are already arriving:
How to use it
Drop one Code node at the end of your workflow. That’s it — about ten minutes, no server, no dependencies. From then on every run leaves a signed receipt.
The n8n walkthroughStart by checking a receipt yourself — one click, nothing uploaded — so you can see exactly what the green seal means before you produce one.
Check a receipt in your browserStraight answers
Try it
That’s the whole idea. Check a real receipt in your browser right now — nothing uploaded, nothing installed. Green seal, genuine record; red seal, it was changed. That’s the whole test.
Prefer the deep end? The full technical account covers the threat model, the two independent verifiers, and the limits.