Custom terms
Project names, clients, systems and codes that only mean something inside your organisation: how to maintain the list, how matching works and where the list is kept.
Bombus recognises data that has the same form everywhere, such as a passport number or a name. It cannot know that "Project Havik" is the code name of your acquisition or that "Atlas" is your client system. Terms like these, which only mean something inside your organisation, go on a list of your own. When someone types or pastes one of them into an AI tool, it is highlighted like any other data type, and once masked it becomes a placeholder that shows its category, such as [ORG-PROJECT1]. The terms work in documents too, following the profile set for Documents.
You find the list under Policy > Detection profiles, on the Custom terms tab.
Custom terms in the dashboardYour organisation's list: add terms, pick a category, delete, import and download as CSV.Transcript
On screen: Heading: "Words that are sensitive only to you". A browser window with ChatGPT. The user wrote: "I need to send an update about the planning." ChatGPT replied: "Sure. What should it say?" Someone types in the text field: "Summarise this for the team: Project Havik is running late, the link with Atlas is not ready yet." Nothing in the text is marked.
A detection model recognises common data such as names and numbers. Project names or the names of internal systems are only sensitive within your organisation, and the model does not know them.
On screen: Heading: "Adding custom terms", with the label "Detection profiles". The window shows the BeeSensible dashboard at app.beesensible.eu/policy/detection-profiles/terms. In the menu, "Detection profiles" is selected under "Policy". The page "Detection profiles" has the tabs "Profiles" and "Custom terms"; "Custom terms" is open. At the top it says: "Add terms that are specific to your organisation. In each detection profile you decide where they are recognised and whether they are standard or critical." The list is still empty. "Project Havik" is typed in the "Term" field, "Project" is chosen as the "Category", and "Add" is clicked. The term then appears in the list with the category Project and the date Oct 5, 2026, and above the list it says "1 term of at most 5000".
You add such words on the Custom terms tab under Detection profiles, where you type the term, choose a category and click Add.
On screen: Heading: "Importing a list", with the label "CSV import". "Import CSV" is clicked. The window "Import terms" explains: "One term per line. A second column with the category is optional. Terms already on the list are skipped. Check the first lines before you import." The file organisation-terms.csv is dropped into the window. It says: "18 terms found in organisation-terms.csv", with a preview of the first five lines, such as Project Havik (Project), Atlas (System) and Kestrel BV (Business relation). After a click on "Import" it says: "17 terms added. 1 was already on the list." The list now shows "18 terms of at most 5000".
If the terms are already in a file, you can import them all at once as a CSV.
On screen: Heading: "The term in an AI tool", with the label "In the extension". Back in ChatGPT the same text is typed again. This time "Project Havik" and "Atlas" get a yellow line underneath, and the BeeSensible tab on the edge shows 2. At "Project Havik" a small window opens with the category "Project", the level "Sensitive", the button "[ORG-PROJECT1]" to mask it, the button "Delete" and the switch "Apply to all occurrences".
When a colleague types or pastes one of these terms into an AI tool, BeeSensible marks it like any other sensitive detail.
On screen: Heading: "What the AI tool sees", with the label "Masking". The mask button is clicked and "Project Havik" becomes "[ORG-PROJECT1]". The same then happens to "Atlas", with the category "System": it becomes "[ORG-SYSTEM1]".
When the term is masked, a placeholder with the category takes its place, so the AI tool still knows what kind of detail was there.
On screen: Heading: "Level and storage", with the label "Detection profile". Back in the dashboard, the sentence at the top of the page is outlined: "In each detection profile you decide where they are recognised and whether they are standard or critical." Above it a padlock appears with "Stored encrypted".
In each detection profile you set whether a category is standard or critical. The list is stored encrypted, and the figures only show the category.
On screen: End card: the BeeSensible logo with the title "Custom terms".
Add terms
To add a single term, fill in Term, choose a Category and click Add. If you already have the terms in a file, Import CSV adds them in one go: one term per line, optionally with the category as a second column, such as Project Havik,Project. Before the import starts, you see how the first lines will appear in the list, and the import skips terms that are already on it. Download CSV exports the whole list, for example so you can update it outside the dashboard.
A term is 3 to 100 characters long, and the list holds up to 5,000 terms.
Category and level
The category you choose for a term sets which placeholder appears after masking. The ORG prefix shows in the prompt that the value came from your own list and was not recognised by Bombus.
| Category | Placeholder |
|---|---|
| Project | [ORG-PROJECT1] |
| Business relation | [ORG-RELATION1] |
| Person | [ORG-PERSON1] |
| Product or service | [ORG-PRODUCT1] |
| System | [ORG-SYSTEM1] |
| Location | [ORG-LOCATION1] |
| Code or reference | [ORG-REFERENCE1] |
| Other | [ORG-TERM1] |
You set whether a category is detected, and how strictly, on the Custom terms tab of the app's detection profile. There you choose Standard, with a yellow highlight, or Critical, with a red one. A category you leave alone is detected at Standard. In the extension, such a highlight carries the label "From your organisation", so the employee can see where it comes from.
How matching works
BeeSensible matches the terms on the list literally. That makes it predictable what a term will highlight, but it also means that variants are only covered if you add them yourself:
- Capitals, accents, spaces and hyphens make no difference:
Project Havikalso findsproject-havik. - A term you enter entirely in capitals only matches capitals, so the abbreviation
KITdoes not highlight the ordinary word "kit". - Typos and compound words do not match, so
Havikdoes not highlightHavikproject. You add such a variant as a term of its own. - When two terms overlap in the text, the longer one gets the highlight.
Where the list is kept
The extension searches for the terms on the employee's device, after detection has returned its result, so the list never goes to the model. It is stored encrypted in BeeSensible's database and on the device, and the extension removes it on sign-out. A change to the list reaches every device within five minutes. The desktop app also searches on the device itself when it scans a document locally.
The figures only show the category, such as "Project (custom term)" under Sensitive data, where the All custom terms filter shows all custom terms together. The audit log records how many terms were added or removed, without naming the terms themselves.
A term is not highlighted
First check in that app's detection profile whether the term's category is switched on. If it is, check that the text contains the term exactly. Compound words and typos do not match, and a term you entered in capitals only matches capitals.
