Security and privacy
The safety rules are the product, not a switch
This page sets out what is built into the software, and what we explicitly do not promise.
A person approves every reply
No reply reaches a customer without a person approving it. Automatic sending is off, and for food safety, payments and privacy it cannot be switched on.
The software only proposes automatic sending for a category once your team demonstrably sends its drafts almost unchanged. Switching it on is always your decision.
Sensitive questions go to a person
Questions about food safety and allergens go to a team member.
Questions about payments and personal data go to a team member.
Before every production release a fixed safety test of 99 questions runs. One misrouted allergen, payment or privacy question stops the build. Measured on 19 October 2026: 99 out of 99 correct.
What it does in your mailbox
It reads your inbox and Sent folder over IMAP. It marks messages as read or flagged, and puts a copy of sent replies in your Sent folder. It does not put drafts or labels in your mailbox: those live in the review screen.
Numbers come from your webshop, not from the AI
Order details and amounts are read from WooCommerce, read-only, and used exactly as WooCommerce provides them.
Personal data and training
Before text goes to the AI model or is used as an example, names and contact details are masked automatically. No model of our own is trained on your data.
Masking works on patterns. That is not a guarantee, which is why a team member always decides before anything reaches a customer.
Photos a customer attaches to a routine question go to the AI model unmasked. For questions about payments, privacy or food safety they do not go to the model.
Customer conversations are deleted automatically after a fixed retention period, 90 days in the current installation. Masked example replies in the knowledge base are kept longer.
Where your data is stored
The application and the database run at Render in the EU, in Frankfurt.
To draft replies, masked text and, for routine questions, attached photos go to Anthropic in the United States, under a data processing agreement with the EU standard contractual clauses. Under Anthropic's commercial terms, input sent through the API is not used to train models.
We do not promise that the AI provider keeps nothing: there is no such agreement.
Sub-processors in the current installation
- Render Services, Inc.: hosting of the application and database, EU region Frankfurt.
- Anthropic, PBC: drafting replies, United States, under the EU standard contractual clauses.
Transparency about AI
Linedesk365 is a tool for your team. The AI suggests drafts and sorts email; a person decides what your customer reads. Someone is always responsible for every reply.
We recommend telling your customers in your privacy statement that you use AI to help draft replies.
Curious what this means for your customer email?
In half an hour we show how it works, starting from your own questions.