AI information
There is no AI in SignaCore.
Not in field detection, not in the signing flow, not anywhere. No part of SignaCore sends a document, a signature, or anything you type to a model. Nothing you upload has ever been used as training data, by us or by anybody else.
Why a page saying what we do not do. Almost every tool you are comparing us against has an AI page describing what its models do with your documents. Ours describes the absence of them, because that is the fact, and because a signing product is a strange place to be relaxed about where your contracts go.
What field detection actually is
SignaCore finds the places to sign in two ways, and neither is a model.
If the PDF carries real form fields, those are read directly out of the file’s own structure. The document already says where its fields are; SignaCore just reads it.
If the PDF is flat, the page is measured. Rules and underlines, the position and wording of captions, the boxes drawn on the page, the shape of a table, the glyphs used for tick boxes, how far a label sits from the space beside it. Those measurements run through rules written by hand and tuned against real documents.
One consequence worth knowing, because it is the practical difference: it is deterministic. The same PDF produces the same fields every time. It does not improve quietly, it does not drift, and it cannot hallucinate a field that is not there. When it is wrong it is wrong in the same way twice, which is why it can be fixed — and why you are shown everything it found before anything is sent to anybody.
Where your documents go
Nowhere, beyond what the job requires. Documents, signature images, and the values recipients enter are encrypted at rest and decrypted briefly inside the service to detect fields, render a page you are authorised to see, produce the completed PDF, and deliver it.
The only outside services SignaCore talks to at all:
- Google and Apple, if you choose to sign in with them. They receive nothing about your documents.
- Stripe, for payment. Card details never reach SignaCore, and Stripe receives nothing about your documents.
- Our email provider, which carries invitations and completed copies. This one does handle document contents, because an attachment is how a completed copy reaches the person who signed it.
- Error reporting, which records that something failed and where in the code. Values are scrubbed before they are sent and personally identifying data is switched off at the source.
No model provider appears on that list, and none is installed in either part of SignaCore. That is checkable rather than asserted: there is no model library in the codebase to call.
If that ever changes
It might. Some of what SignaCore does by hand would be better done by a model, and we are not promising never to use one. What we are promising is the order of events.
Before any customer document reaches any model, this page will say which feature uses it, which provider runs it, what is sent, what that provider is contractually allowed to do with it, how long it is kept, and how to turn the feature off. Published first, used second.
Documents will not be used to train models — ours or anybody else’s. That one is not conditional.
Questions this page should answer
Is my document used to train anything? No. Nothing you upload has ever been used as training data.
Does detection get better by learning from my documents? No, and that is deliberate. It improves when somebody changes the rules, which is visible in the code and testable against documents we hold ourselves.
Can I opt out of AI features? There is nothing to opt out of. If that changes, opting out will exist before the feature does.
Is anything about my documents sold or profiled? No. Document contents are not sold, not used for advertising, and not used to build a profile of anybody.
Last reviewed 7 October 2026 against the code as it stands. The security boundary this sits inside, and the parts of it we deliberately do not claim, are on Trust & data.