SourceSecure by Jacob

SourceSecure stops client details reaching ChatGPT, Claude, Microsoft Copilot, Google Gemini and the other AI chat tools it covers, by accident.

It reads what you are about to send, finds the things that should not leave your firm, and offers you a version with those details replaced. You choose whether to send the redacted version or what you typed. If its own check cannot finish, it holds the message until it has been checked rather than letting it go unchecked.

WHAT IT CATCHES

Client names, account numbers, SINs and SSNs, card numbers, addresses, postal codes, email addresses, phone numbers, company names and places. Your firm can add its own terms as well, such as a fund name or a project codename.

IT READS YOUR ATTACHMENTS TOO

PDFs, Word documents, spreadsheets, PowerPoint decks, plain text and scanned images. AI sites upload a file the moment it is attached, so SourceSecure reads it first, before the site gets it. A clean file goes straight on. One with client details in it waits: it shows you what it found, page by page, and you can black out anything else by dragging over it or undo a redaction with a click. Then you attach a redacted copy, or the original if you choose.

IT ALSO CLEANS WHAT YOU CANNOT SEE

Author names, tracked changes that were deleted but never accepted, comment authors and document properties all travel inside an Office file. SourceSecure removes them from the copy it attaches.

EVERYTHING HAPPENS ON YOUR COMPUTER

The model that recognises names and places is a file inside this extension, not a service somewhere. Your messages and your documents are read and checked in your own browser. Documents are never sent to us, and no message text is sent until it has been redacted there, and then only for the optional features described below. Checking needs no network: switch it off and messages are still checked.

FOR TEAMS

An admin can see whether the firm is actually using it, set terms everyone must always redact, and require that the high risk kinds stay switched on for everybody. They can see figures for the whole firm or one person, and each person's catches: the kind of detail, the site, what was done, and the detail masked on the person's own computer first (like •••• 4242), and when protection was paused or switched off. No message text is ever stored by us or shown to an admin.

FOUR KINDS OF SEAT

A firm buys seats and gives each person one. Guard is the redaction above. Guard Pro adds prompt coaching. Guard 360 adds reply marks. Guard Enterprise, bought as a team, adds the Living Vault and a shared firm playbook. What a person's browser does depends on their own seat.

PROMPT COACHING, ON GUARD PRO SEATS AND ABOVE

Guard Pro seats and above can suggest how to ask a better question. That needs a model we run, so a redacted version of the conversation is sent to it and on to OpenAI, with the client details found already replaced by stand ins such as [PERSON_1]. Help me reads the whole conversation on the page, including anything typed but not yet sent, and starts as soon as a tip appears so it is ready when clicked. Sharpen my prompt starts the same way when the pointer rests on it for a moment, which sends the redacted conversation then, before the click. We store none of it, and it is sent to OpenAI with its storage option off. Any person can switch prompting tips off from the extension's icon, and with it off coaching sends nothing.

Once a week it also writes each person a short private note about how their week with AI went, compared only with their own earlier weeks, with one habit to try next. The note is written from counts and, on Guard 360 seats and above, a few short facts about their work from their vault, never from the messages themselves, and only they can see it. Their firm's admin can see which habit each person is working on, never the note.

REPLY MARKS, ON GUARD 360 SEATS AND ABOVE

When an AI reply finishes, SourceSecure faintly underlines the places worth a second look: a decision made for you, an assumption, something stated too surely, or something that contradicts what your firm has said. Point at one to see why, and ask a follow-up it drafts into the message box. Nothing is ever sent for you.

To do that, your message, the AI's reply and the few messages before them are redacted on your computer and sent through our server to OpenAI, with the facts in your vault. The vault is short facts about your work and your firm, learned from your messages or saved by you, kept in this browser only, and yours to see, change or delete. We store nothing from a reply check. Reply marks have their own switch, and switching coaching off stops them too.

THE LIVING VAULT, ON GUARD ENTERPRISE SEATS

A private memory of your work that stays on your computer, encrypted. It keeps what you choose from your AI chats (a short summary, decisions and deadlines), notes you write, files and folders you add, meeting transcripts from Zoom, Teams and Google Meet, and anything you capture from a page with the right-click menu. It builds a page for each client and company you work with, with everything known about them, and a map of how they connect.

Ask it in plain English, like "did we ever finish that letter for the Chens?", and it answers from your own notes, with links to where each part came from and to the original chat. On an AI chat page, Add notes finds the notes that bear on what you are writing and, when you press Write background, writes a few sentences of background from the ones you tick for you to change and add; or it adds the notes as they are, which sends nothing. The names and private details SourceSecure finds are swapped for stand-ins before anything is sent to answer a question, read a chat or a meeting, or write a background, and the real names are put back on your screen. A question is sent only when you ask it, and not while SourceSecure is paused; a meeting transcript is sent to be summarised when you add it. Nothing from the vault is stored on our servers. It opens in the browser's side panel beside any page, from the pill on an AI chat page, the toolbar icon, or a keyboard shortcut.

Team heads write the firm's playbook, its rules and approved wording, which every Enterprise seat uses and which reply marks check answers against. The playbook is the one shared part, and it refuses anything that looks like client details.

REQUIRES A SUBSCRIPTION

SourceSecure is sold to firms. Sign in with the address your firm added. Pricing and a free look at what it does are at sourcesecure.ca.

Privacy: sourcesecure.ca/privacy
Terms: sourcesecure.ca/terms
Security: sourcesecure.ca/security