Schema Builder is where you define what data you want Docxster to pull out of a document. A schema is really just a list of fields, invoice number, total amount, vendor name, that kind of thing, and until now every one of those fields had to come from somewhere in the document itself. The model had to find it, read it, and map it.

That's fine when the value actually lives in the document. It falls apart the moment you need a field that doesn't, a value you already know before you've even looked at the file.

It's a toggle, not a separate field type. Turn on "Use Constant Value" for any field, of any data type, and enter the value once. That field will always return the value you entered, regardless of what the document actually contains, and because there's nothing to look for, it's never part of what gets sent to the model.

Where this fits

Say every invoice you process needs to carry an internal cost center code, or a department tag, or a fixed processing code your downstream system expects, none of which ever show up on the invoice itself. Before, you'd either bolt that value on after extraction in a separate step, or hope the model didn't try to guess at something that wasn't there.

Now you flip on Use Constant Value for that field once, and every extraction under that document type carries it automatically. No prompt tuning, no post processing step, and nothing for the AI to get wrong, because it never has to look for it in the first place. If you've got predefined values that have nothing to do with the document content, this is where they belong now, not in the prompt, not in a downstream script, just set once and done.

How it works

1. Go to Workflows, then Document types, and open a document type's Smart Document Setup.

2. Add a new field or edit an existing one. Pick its data type as usual (Number, String, Date, whatever fits).

3. Turn on Use Constant Value and enter the value in place of an extraction prompt.

4. Hit Update Field. Every document run through Smart Document Setup for that type now returns this value for the field, no extraction involved.

Small feature, but it removes a whole class of workaround: no more downstream scripts bolting on a code the document never had, and no more fields that only exist to hold a value the AI was never going to find anyway.