Security & Privacy
Data privacy
Who processes your data, what Agent Hero staff can see, and the data policy that hides sensitive numbers.
Your data
Your documents and chats belong to your organisation. Model providers process each turn to answer it. They don't get your workspace, your document store, or anything the agent couldn't see in that turn. Web searches go to the search service.
This model is cheap because its provider may use what you send to improve its products. Don't use it for confidential or personal data, and see Models.
Google data is used only for the features you connected and shared only with the contracted processors that run them. It is never sold, used for ads or credit decisions, or used to train generalised AI models. Agent Hero shows this before you connect.
Plato (Read Only) reads lists and reports only, with clinical fields removed before the agent sees anything.
Agent Hero staff access
Agent Hero staff can open your organisation in support mode to help you. They see its agents, documents and connections read-only, and can chat with an agent at view-only access. Those turns are billed to Agent Hero, not to you.
Staff can also preview the app as one member sees it, including that member's conversations, read-only. You are not notified. Agent Hero logs every support session and preview.
Deleting and erasing
Deleting a document removes it from your workspace and search, but doesn't yet erase every stored copy. Ask your Agent Hero contact to erase content, or about data residency, retention and DPAs, which vary by deployment.
The data policy
A data policy hides sensitive numbers, such as NRIC numbers and bank accounts, in what people see in chat and search. An admin sets it once, and it covers every agent in the organisation.
It is built but not switched on for any organisation yet. The rest of this page describes how it will work once it is.
What it hides
| Detail | Choices | Starts at |
|---|---|---|
| NRIC / FIN | Show last 4, Hidden, Show in full | Show last 4 |
| Card number | Show last 4, Hidden, Show in full | Show last 4 |
| Bank account | Show last 4, Hidden, Show in full | Show last 4 |
| Mobile number | Show last 4, Hidden, Show in full | Show last 4 |
| Salary | Hidden, Show in full | Hidden |
| CVV | Always hidden | Hidden |
Show last 4 keeps the end of the number, such as •••• 1111. Words around it, such as "DBS" or "salary", stay visible.
Detection follows fixed rules, not AI. It looks for a number's shape, its check digits, or a label nearby. It is built for Singapore numbers, such as NRIC, FIN, local mobiles and local bank accounts. A salary is caught only next to a word like "salary" or "gross". Names, addresses and health details aren't covered.
Where it applies
The policy hides matching numbers in:
- Replies as they stream into chat, tool results included
- Replies sent to Telegram, and to Slack once it is available
- The saved chat history
- Search results and previews
- Values the agent suggests in a form it asks you to fill
It follows the agent's organisation, not the viewer's.
The agent still reads full values in the documents it opens, and anything it sends, such as an email or a Xero entry, uses them. Approval cards and the original files also show full values.
Also not covered:
- What you type yourself
- Documents opened or downloaded by anyone with access
- Document titles outside search
- Your documents read by another organisation's agents
Set the policy
- 1
Open Data protection
Go to Settings and choose Data protection under Organization.
- 2
Turn it on
Switch on Protect sensitive data. It applies to every agent in your organisation.
- 3
Pick a choice for each detail
Each change saves at once.
Search updates about a minute after your last change, and a line under the list shows its progress. Turning the policy off brings full values back into search the same way.
Only organisation admins see the page in Settings and can change it. Members who open it see it read-only. Agent Hero staff can't change it from a support session. Every change is recorded with who made it and when.
A masked copy of an Excel file
With the policy on, an agent can make a masked copy of an uploaded CSV or Excel file:
Make a masked copy of @Payroll.
The copy is a new Excel file saved with the agent's files, titled with "(masked)". Every cell is stored as text. The agent checks each cell after writing, and saves nothing if one doesn't match. It handles up to 10,000 rows and 200 columns.

