The animated scene
Follow the request all the way to the model.
Go through the scene at your own pace or replay a step: in this fictional example, three detected values give way to placeholders before sending.
Fictional data; the model shown does not reflect the catalog of your plan.
Step by step
The request keeps its meaning. The detected values are replaced.
Drafting a payment reminder doesn’t require sending your client’s bank details to the model. In this fictional example, four values give way to placeholders, then go back in place in the answer.
Original message: Draft a reminder for invoice F-2026-118 issued to Atelier Brunet, 15 days overdue. Include the IBAN FR76 3000 6000 0112 3456 789 and the SIRET 812 345 678 00044. Version sent to the model: Draft a reminder for invoice FACTURE_01 issued to ENTREPRISE_01, 15 days overdue. Include the IBAN IBAN_01 and the SIRET SIRET_01.
Restored answer: Hello, according to our records, invoice F-2026-118 issued to Atelier Brunet is still outstanding. You can pay by bank transfer to the IBAN FR76 3000 6000 0112 3456 789.
See the fictional matches
- Invoice
- F-2026-118FACTURE_01
- Business
- Atelier BrunetENTREPRISE_01
- IBAN
- FR76 3000 6000 0112 3456 789IBAN_01
- SIRET
- 812,345,678 00044SIRET_01
- DetectMoon Blur scans the message and underlines the values it recognizes.
- MaskEach detected value gives way to a placeholder, such as FACTURE_01.
- SendThe model receives the request with the placeholders, without the detected values.
- RestoreBack in your workspace, the values return to their place in the answer.
What it detects
From contact details to business data.
The 151 categories of the catalog cover the families below, among others. A supported category does not guarantee that every value will be found.
- 151categories of data in the detection catalog
- 20families, from technical secrets to health data
- 100terms of your own to mask from Constellation, up to 1,000 with Univers
Identity and contact details
Email addresses, phone numbers, passports, social security numbers and bank details.
Fictional example · Email addresscamille.martin@example.com becomes EMAIL_01.Legal and health
RG docket numbers, appeals, RPVA, ECLI, RPPS and FINESS identifiers, IPP numbers and medical record references.
Fictional example · RG docket numberRG 24/01234 becomes RG_01.Business and finance
SIRET, European VAT numbers, DUNS, LEI, invoices and SEPA references.
Fictional example · SIRET812,345,678 00044 becomes SIRET_01.Technical secrets
SSH or PGP keys, JWTs, API keys and database URLs.
Fictional example · Database URLpostgres://app@db.example.com/prod becomes URL_BDD_01.
These figures describe the detection catalog, not a detection rate.
Stay in control
Control starts before you send.
Detection can miss things. Sensitive information can also be inferred from context, even when some fields are masked.
Before you send
- Leave out details your request does not need.
- Check what you are allowed to share.
- Avoid secrets, access credentials and unnecessary documents.
- Review the answer before you use it in your work.
The real values only come back in the answer you read: the model itself only saw the placeholders.
Understand the whole chain
- Your messageYou write your request in your Moon workspace.
- Moon BlurDetected values give way to placeholders before sending.
- Model providerIt receives the masked version. Depending on the model, it may operate outside the EU.
- Your answerThe matching values can be put back in your workspace.
Moon AI is hosted in France, in the European Union. Depending on the models and features you use, some providers operate outside the EU: the DPA lists them and the framework that applies.
Moon CLI and API
In your terminal and your apps.
Moon CLI
Moon CLI masks the data it detects on your own machine, before anything is exchanged with the model.
API Moon Blur
A Moon Blur API to mask the data in your documents and data flows.
Coming soonRead the docs
Get started
Moon Blur is included from the Free plan.
Create an account with no credit card: Moon Blur masks the sensitive data it detects in your conversations from your very first message. For an organization, decide first which data, tools and responsibilities are involved; the studio can help you frame it.
