1. Where you meet AI in Verimly
- The assistant in the back office. A chat panel where you can ask questions about your business, find help, and get changes prepared for you. It is an AI system (a large language model), not a person. The panel says so the first time you open it and shows it in every conversation.
- Agent access (MCP). If an owner or manager connects an AI agent such as Claude, ChatGPT, Cursor or Codex to Verimly, that agent can use the same tools as the assistant, with the permissions of the person who connected it. The agent and its provider are chosen by you, not by Verimly.
- Search in the help centre. Help articles and your question are turned into numeric vectors to find the best article. This is part of how the assistant answers.
- Nothing else. Verimly does not use AI to decide about staff, guests or customers, to score people, or to read faces, voices or emotions. Tip scores and similar calculations follow fixed rules described in our other notices.
2. What the AI can and cannot do
- It can only see what you are allowed to see. It uses your own role and permissions. If you cannot open a report in the back office, neither can the assistant for you.
- It prepares, you decide. When the assistant wants to change something, such as a price, a menu item or a setting, it shows you exactly what will change. Nothing happens until you confirm.
- It never makes refunds, voids or payments. It can explain how to do them and open the right screen; a person does it there.
- It can be wrong. Language models sometimes give answers that sound sure but are incorrect, out of date or incomplete, especially for numbers, tax and legal questions. Check anything important before you act on it. The assistant is not an accountant, lawyer or adviser.
- It is not for decisions about people. Do not use it to decide about hiring, firing, disciplining or paying staff, or about individual guests.
3. Who processes your data
Your questions, the data the assistant looks up for you, and its answers are processed by Anthropic (the language model) and, for help search, Voyage AI. Both are sub-processors of Verimly and are listed on the Subprocessors page. Anthropic processes data in the United States or other regions of its global service, so your data leaves the EU; we rely on Standard Contractual Clauses. Under its commercial terms Anthropic does not train its models on this data and deletes it within 30 days, unless a request is flagged for violating its usage policy. We configure Voyage AI so that its models are not trained on our content.
Files you attach to a message (photos, PDFs, spreadsheets) are sent to Anthropic with your message so the assistant can read them, for example to import a menu. Their content is not masked, so attach only what is needed. Only you can open your attachments; they are kept with the conversation for 90 days and deleted earlier when you delete the conversation. Text inside a file is treated as data, never as instructions to the assistant.
Before data reaches the model we mask guests’ phone numbers, email addresses and addresses, and we never send card details. Analysis tools return totals instead of individual people wherever possible.
We keep your conversations for 90 days (you can delete them yourself earlier) and a record of each tool call for 1 year. Details are in the Privacy Policy, section 5, and in the Data Processing Agreement for data about your guests and staff.
4. Your choices
- The assistant is optional. You can run Verimly without ever opening it, and you can delete your conversations from the panel.
- An owner can switch off the assistant, or agent access, or both, for the organisation under Settings, Agents and MCP. The owner can also require read-only access and revoke any connected client or agent key at any time.
- If you connect your own agent, start with read-only access, and keep keys secret. What your agent’s provider does with the data is governed by your agreement with that provider.
- Your data protection rights (access, correction, erasure, objection and more) apply to the data in conversations and logs. See the Privacy Policy, section 9.
5. Feedback and reporting problems
Every answer in the panel has a thumbs up and thumbs down so you can tell us when it was wrong or unhelpful; we use this to improve the help content and the instructions we give the assistant, not to train a model. If the assistant gave harmful, discriminatory or clearly incorrect output, or you think someone misused agent access, email hello@verimly.com with the subject "AI issue" and, if you can, the time and organisation. For security problems, use security@verimly.com. For privacy questions, write to privacy@verimly.com.
6. How we manage the risks
- Permissions are enforced by the system, not by the model, so the assistant cannot be talked into showing something you are not allowed to see.
- Every change is a proposal that a person confirms, with an expiry; money-moving actions are not available at all.
- Text from guests, such as notes on an order, is treated as data and never as instructions to the assistant.
- We test the assistant on realistic questions before releasing changes, log tool calls and watch for abuse and errors.
- We keep the models replaceable. If we change a provider that processes personal data, we give the notice described in the Data Processing Agreement.
7. Changes
We update this notice when AI features or providers change. The date at the top shows the latest change. The AI features are also covered by the Terms of Service (section 23) and the Acceptable Use Policy.