SourceLace Docs
FAQOpen the app

Data protection

Choose what the AI may see of the values SourceLace reads from your systems: hide fields, mask values such as email addresses and card numbers, and still let the right people see and use the real values. Admins set this on the Data protection page. The one exception is a photo: if your admin turns photos on, the AI reads the picture, and data protection rules cannot check what is in it (see Attaching photos).

How it works

Your rules apply to everything that reaches an AI through SourceLace: the assistant in the SourceLace app, AI agents, and your own AI app (over MCP) if you use one. They add to what SourceLace already does: each person uses their own login, access by group narrows what they reach, results are never copied to a database, and every call is audited.

  • Hidden fields never reach the AI. They are left out of every result.
  • Masked values reach the AI as placeholders: short labels in place of the real values. The same value gets the same placeholder throughout one chat, agent or AI app, so the AI can still say "the same person" and work with it. Another chat gets different placeholders.
  • The person who asked sees the real values in the SourceLace app, and a change, draft or message the AI proposes is made with the real values, after the person has confirmed the preview. The AI never sees them.

Everything is off until an admin turns it on.

Field rules

A rule names a source, an object (or * for every object) and a field, and says what to do with it:

Rule What the AI gets
Hide Nothing: the field never reaches the AI.
Mask A placeholder instead of the value.
Allow The value as it is, even when a detector would match it (for example a shared support address).

For a Marketing Cloud data extension, write the rule with its key (the External Key shown in Marketing Cloud); if you type the name people see in the list, SourceLace saves the rule under the key, or asks you for the key when it cannot tell which data extension you mean. Field names are not case-sensitive. A rule for a table or object also applies when someone types another spelling of its name, for example with or without the schema; where SourceLace cannot tell whether two spellings are one object, the rule applies to both (see Access by group). SourceLace checks every query so that hidden fields stay hidden and masked fields stay masked. A rule can list groups whose members see the real value. For them a hidden field is masked instead of removed, and a masked value is shown to them in SourceLace and can be used in changes. Without groups, everyone who can read the record sees the real value in SourceLace. The AI only ever gets the placeholder.

When a question asks for a hidden field

If the assistant asks for a field your admin hid (say you ask for email and the rule hides it), the chat shows a step called Protected field left out, marked with an "i" and not as a failure. Open the step to see which field it was. The assistant then asks again without that field and says in one sentence in its answer that a protected field was left out. Nothing hidden is shown. If you need the field, ask your admin to change the rule or to add your group to it.

Detectors and your own patterns

Detectors find values inside any text, such as notes, email bodies, chat messages and file text, and mask them. This includes a question a person types in the SourceLace chat: an email address typed into a question reaches the AI masked, and the person still sees what they typed. Turn on the ones you need:

  • Email addresses
  • Phone numbers (written with separators or a leading +, such as +44 20 7946 0958 or (415) 555-0134)
  • US Social Security numbers (such as 123-45-6789)
  • Payment card numbers
  • IBANs (international bank account numbers)

Add your own patterns for identifiers such as employee numbers, as regular expressions (for example EMP-\d{6}). SourceLace checks your patterns and refuses ones it cannot run safely.

Test a sample

Test a sample on the page shows exactly what the AI would get for one record, so you can check your rules on real data:

  1. Choose the source. If you have not signed in to it yet, or SourceLace cannot read a record from that kind of source, the page says so right under the box, and you can paste a row instead. Changing the source clears the object and the record, and lists the new source's objects.
  2. Choose the object from the list of that source's real objects (you only see the ones your own access reaches; the box starts empty, so click it to see the whole list, and type to narrow it). Leave it empty (or *) only when you paste a row, which then applies to any object.
  3. Press Use a real record. SourceLace reads one real record of that object with your own sign-in and your own access, shows it in the Record box, and shows underneath what the AI would get: which fields are hidden and which values are masked.

The record is shown only to you, in the page. It is not stored and not written to the audit trail (the audit trail notes that a record was read, or that an attempt was refused, from which source and object, and how many fields, never a value). A long value is cut short, at most 80 fields are read, and the whole record is kept to a size the test accepts; the page says when some of the object's fields are not in it. Fields your rules hide are still shown to you in the Record box, because that is the record as the source has it; the box underneath is what the AI gets.

If your groups have a read rule that narrows which rows of the object you may see, SourceLace does not read a sample record of it, because the first record could be one you may not read: the page says so. The same goes for a kind of source SourceLace cannot read a sample from (health records are never read). In both cases you can paste one row as JSON into the Record box instead and press Show what the AI sees. Opening the page lists the first source's objects, which is recorded in the audit trail like any search of a source's objects.

Real values for the right people

  • In the SourceLace app answers, tables, charts and change cards show real values to the person who asked, if they may see them. Saved chats show real values again when you reopen them, and are deleted according to your retention settings.
  • Changes, email drafts and chat messages the AI proposes are made with the real values, only for people who may see them. If that cannot be done safely, nothing is changed.
  • In your AI app (over MCP) the AI and the previews it shows you have placeholders only. Open the SourceLace app to see real values.
  • AI agents work with placeholders too; their drafts and changes are made with the real values, and the agent's results chat shows them to its owner.

In the assistant's written replies, a real value that SourceLace swaps in for you is marked with a small highlight and the words "shown to you only" (hover over it, or hear it with a screen reader). It tells you the value is on your screen because you may see it, and that the AI only had a placeholder. Copying the text copies the value, so share it with care. Someone who may not see the value sees the placeholder instead, with no highlight.

Nothing that stands behind a placeholder is logged or written to the audit trail, and it is deleted with the chat or at the end of your organization's result time limit (see Data retention controls).

What is not covered

  • What can be worked out from other information: protection cannot cover values the AI already knows, what someone can work out from other fields they have, or a person's own wording (a customer's name, a project name) when no detector or pattern recognises it. Hide a field that must not be reasoned about at all. Field rules (hide or mask a field) work on what SourceLace reads from your systems, so they cannot apply to words someone types; the detectors and your own patterns do.
  • Data inside images, scans and attachments that SourceLace does not read as text, and the contents of photos (the AI reads a photo if your admin turns photos on), and values written in formats the detectors do not recognise.
  • Field names and record ids are never masked.
  • Documents and spreadsheets the assistant makes (Word, PowerPoint, PDF, Excel, CSV) show placeholders, and leave hidden fields out.

When something goes wrong

What you see What it means What to do
"That table name has spaces, quote marks or invisible characters in a place where it cannot be read as written. Remove them and ask again." A rule was saved on, or a question named, a table whose name has spaces, quote marks or invisible characters at an edge or around one of its parts. A data protection rule cannot be saved on such a name, and a question that names one is refused before anything is read. Retype the name without the extra characters (copying from a document often adds them), then save the rule or ask again. A space inside a name is fine.

Audit

Every rule change is in the audit trail. Each call's entry records how many fields were hidden and values masked, never the values themselves.

A SourceLace Runner applies the same rules.

More short answers: Security and data protection in the FAQ.