Common questions
Where is the data stored?
In Amazon Web Services’ London region (eu-west-2). Pilots are hosted and operated by us, on a server of its own for each customer, shared with no other customer. The pilot’s data stays in London: the database and invoice files on that server’s encrypted disk, their backups, the server’s logs and its secrets. So do our records of each operator’s session on the server, which can name a reviewer’s account, and our copy of the record of each change to its hosting account. A few things are outside London: the uptime checks AWS runs against the server from other regions, the alerts they raise, and the server’s web addresses, which are public, none of which holds a payment or an invoice; and the rest of our own records of the work we do on each deployment, kept on our own computer and phone: the deployment’s record, which names your reviewers’ accounts, and the mobile numbers you give us for your reviewers and your security contact. Our emails to you go through Google Workspace, and reviewers’ passwords are sent by text message, as our privacy notice says. At the end of the pilot, its data is returned or deleted, as the agreement sets out.
The deployment holds the payment details your integration sends, the invoice data you supply for the comparison, the invoice file it fetched, your reviewers’ accounts and sign-in records, each reviewer’s catch-and-miss counts, the decisions your reviewers made, and the signed records. It holds no funds and connects to no bank.
What does a pilot actually involve?
Sixty days. You point a gated tool at the SDK, we run a deployment for you, and your reviewers use the console. The first pilots run in shadow mode only: an approval moves no money, and payments carry on through your existing process. Pilots on a test account, or under a payment ceiling enforced by your payment system, come later.
You supply the reviewers and the invoice data the comparison is made against. We do not review your payments.
Keeping your payment credentials out of your agent’s reach stays with you: a payment credential your agent can reach by any route other than the gated payment tool is outside SignetGate’s protection. Our list of limitations says what else stays with you.
How is this different from an approval step in Slack or our ERP?
An approval step records that somebody clicked. This records what they were given to decide on: the record is signed over a fingerprint of the payment and a fingerprint of the exact invoice file that was put in front of the reviewer, so a later argument about which version of an invoice was on screen is settled by comparing fingerprints rather than by memory.
Two other differences matter in practice. The checks that had to pass are part of the signed record rather than a separate audit log, and your auditor can verify a record without us, including after a dispute has started and without asking us for anything but the public key you already kept.
Why do the package names say signetgate?
Signetry is the company; SignetGate is the product your agent integrates with, and the name on the certificate your reviewers hand over. The letterhead and the imports are naming two different things.
The names an integration depends on are frozen, whatever the company is called: the package names, the environment variables, the key endpoint and the error codes stay signetgate. Renaming them would be a breaking change to every deployment for a cosmetic reason. The signed record itself carries none of those names, so a rename would not touch a record you already hold.
What happens to our evidence if you disappear?
The records stay verifiable. They are signed with Ed25519 over a form anyone can reproduce, and the checker is a single page that works offline against a public key. Keep your own copy of four things from the day you start: the public key, your records, the checker page and the command-line checker that ships beside it. None of them depends on us existing.
What would be lost is the service. Gated payments would not go through: the software’s rule on any failure is to refuse the payment rather than release it, so an agent calling a gated tool would be refused rather than paying unreviewed. Paying through that tool again would mean taking the gate out, which is a decision someone at your end makes deliberately. That is a continuity question for the agreement, not a cryptographic one.
Which browsers does the checker need?
The checker needs a current Chrome, Edge, Firefox or Safari; an older browser is told it cannot run the check, and is never told that a genuine record is bad.
The checker is tested automatically in Chromium (the engine behind Chrome and Edge) and Firefox. Safari is checked by hand on a Mac: the current version was checked in Safari 18.6 on 7 October 2026, with every case matching.