Get started

shim checks the prompts people type into ChatGPT, Claude and Gemini before they are sent, and the files they attach before the site can upload them. A managed extension runs your organization’s rules inside Chrome and Edge. A console that you host keeps the policy, the enrolled browsers and the evidence for every prompt and file that matched a rule.

What happens to a prompt

1. An employee writes in a recognized ChatGPT, Claude or Gemini composer and sends.
2. The extension checks the text on the device against your current policy. Nothing is uploaded to make that decision.
3. If no rule matches, the prompt is sent. The extension reports metadata such as the application and policy version, never the text.
4. If a rule matches, the rule’s action applies: the prompt is sent and recorded, or a review opens where the employee chooses, must edit or mask the draft, or cannot send the original.
5. The console receives the request with the employee’s choice. Administrators can open the original prompt and any applied edit or mask for 90 days after each is captured.

The two parts

Part	Who uses it	What it does
Console	Administrators	Publishes the policy, creates enrollment codes, shows devices and requests
Extension	Employees	Checks prompts and files locally, shows the review dialog, reports matched requests

The console is one Bun server backed by PostgreSQL. It serves the administrator interface and the API that enrolled browsers call. The extension sends its reports only to the console origin you configure.

Set up in this order

1. Deploy the console and create the first administrator.
2. Review the default policy. New organizations start with credentials, email addresses, payment card numbers, IBANs and national ID numbers set to let the employee choose, and with AI app discovery on.
3. Install the extension on managed Chrome or Edge browsers and set the platform URL.
4. Enroll each browser with its own one-time code.
5. Send a test prompt containing an email address and open the request in Requests.

What employees see

When a prompt matches a rule that lets the employee choose, a review dialog opens over the composer with three choices:

• Edit or mask opens the draft beside a masked preview. Send edited prompt or Mask and send sends the result after it passes a fresh check.
• Send once sends the original prompt one time, with an optional explanation that only administrators can read.
• Cancel closes the dialog and keeps the draft.

The dialog tells employees what administrators can review: the original prompt, any applied edit or mask, and their choice. Prompt text is deleted 90 days after it is captured. See Requests and evidence for what is kept longer.