Moon Blur is always on, in every conversation and on every plan. It has no settings, apart from the optional list of your own data to mask. It recognizes contact details (email address, phone number, postal address), ID numbers, bank details, technical secrets and many business references.
How it worksLink to the How it works section
- You write your message and send it.
- Before it is sent, Moon AI looks in the text for the sensitive data it knows how to recognize, and for the terms on your personal list if you have one (Your own data to mask).
- Each value found is replaced with a placeholder that shows its type and a number:
EMAIL_01,TEL_01,IBAN_01. The model receives the message with these placeholders, without the values. - The model answers using the placeholders. Moon AI puts the real values back in place before showing you the answer.
Fictional example:
| What you write | What the model receives |
|---|---|
| Hi, I’m Camille Martin, you can reach me at 06 39 98 12 34 or camille.martin@example.com. My IBAN: FR76 3000 6000 0112 3456 7890 189. | Hi, I’m Camille Martin, you can reach me at TEL_01 or EMAIL_01. My IBAN: IBAN_01. |
In this example, the name “Camille Martin” goes through as is: Moon Blur does not detect names of people, clients or projects on their own. Add the ones you want masked to your list.
The same value keeps the same placeholder for the whole conversation, so the model can follow the thread: if it writes “I’ll call you at TEL_01”, you read the real number.
What goes through Moon BlurLink to the What goes through Moon Blur section
| Content | Masked before it reaches the model |
|---|---|
| Your messages and the whole conversation history | Yes |
| The text of the documents you attach (PDF, Word, spreadsheets…), including a long pasted text | Yes |
| Tool results sent back to the model (pages read, search results, output from the isolated computer) | Yes |
| Background tasks, such as the conversation title | Yes |
| Conversations of agents, routines and connected messaging apps | Yes |
| Images, photos, voice, and data sent to your connected services | No: see the limits |
The list of recognized data types is in Detected data.
Where to see Moon Blur in the appLink to the Where to see Moon Blur in the app section
- Below the message box: on a computer, the label “Protected by Moon Blur”; its tooltip says “Personal data is replaced with tokens before your message reaches the model.” On a phone, only the shield is shown.
- At the top of a saved conversation: the “Moon Blur · 5 items masked” pill counts the items masked in the whole conversation (an item that appears in several messages counts once).
- Below each message you send, when Moon Blur masked something: a note, described below.
- In Settings: “Settings” › “Data and privacy” › “Moon Blur” shows “Always on” and holds your list of terms (open the page).
- Before you publish an artifact through a public link: Moon Blur scans the page and flags the personal data it seems to contain. The published page keeps the real values: it is up to you to confirm that it can be public (Publish an artifact).
See what was masked in a messageLink to the See what was masked in a message section
Below a message, the note “Moon Blur masked 3 items before sending to the model · Show” tells you how many items were replaced.
- Click “Show”.
- The “What the model received instead” panel lists each placeholder (for example
EMAIL_01) next to the original value, partly hidden. - “Show the message sent to the model” displays the whole message as the model read it, with placeholders instead of values.
- “Add a term to mask” opens your personal list, so that what Moon Blur missed is masked next time.
For a message written by someone else (a conversation shared by a colleague, for example), the panel shows the type of each item, never its value.
The badge changes with the result of the scan:
| What you see | What it means |
|---|---|
| “Moon Blur masked N items before sending to the model” | Data from your message was replaced. |
| “… in the conversation context” or “Moon Blur · context protected” | Nothing in this message, but data from the history, attached files or tool results was masked. |
| “partial analysis”, in amber, with “Very long message. Only the beginning was analyzed.” | The text was longer than what Moon Blur scans in one go. See the limits. |
| “An earlier message in this conversation was only partially analyzed.”, in amber | An older message in the history was too long. The “New chat” button lets you start again from a short history. |
| “Moon Blur · sent without protection”, in red | The scan failed and an emergency setting of the platform let the message go. By default, a message that Moon Blur cannot scan is not sent. |
Moon Blur in the API and in Moon CLILink to the Moon Blur in the API and in Moon CLI section
The developer API and Moon CLI handle Moon Blur their own way, and do not use your personal list:
- developer API: Privacy and Moon Blur;
- Moon CLI, which masks data on your computer: Moon Blur in Moon CLI.
In this section
- Detected dataThe families of sensitive data Moon Blur can recognize, how it spots them and the placeholders that replace them, with fictional examples.
- Your data to maskAdd your own terms (client names, project names, internal codes) so that Moon Blur always replaces them with a placeholder before sending to the model.
- LimitsWhat Moon Blur doesn’t mask, why its detection can miss things, and what happens when the analysis of a message fails or only covers part of it.
See alsoLink to the See also section
- Detected data: the data types recognized and their placeholders.
- Your own data to mask: also mask your client and project names.
- What Moon Blur does not cover.
- Moon Blur and sensitive data, the overview on the site.
- Security and commitments.