Security and privacy
How carefully the chatbot handles data and actions, who it trusts, what it remembers and how long conversations are kept.
Open the chatbot's Security.
Security mode
| Mode | What it means |
|---|---|
| Organisation | Follows your company's policy in the Security center. |
| Standard | The default balance. |
| Strict | Only AI providers that don't train on your data; no semantic cache. |
| Confidential | Zero retention, no content logs, no human access to content, no semantic cache, strict data-loss prevention, EU providers only. |
Chat Mode or Agent Mode
Chat Mode answers, finds and recommends. Agent Mode may also create, update, book or cancel, as far as the tool firewall and each tool's run setting allow.
Signed visitor identity
Let the chatbot use customer tools (orders, invoices, subscriptions) only for visitors your own backend vouches for. Create the secret here and follow Visitor identity. Without a secret every visitor is anonymous to the chatbot.
Data retention
How long conversation text is kept. It's erased from the database, not hidden; counts, tokens and costs stay for analytics.
| Setting | What happens |
|---|---|
| Keep for N days | Kept for the number of days you choose, encrypted with your organisation's key. |
| Zero retention | The text is kept only while the chat is going on and erased 30 minutes after it ends; tool data isn't stored. |
What the chat may use
| Setting | |
|---|---|
| Page context | Which page details the widget may send. Read headings and product data is opt-in. Form fields, passwords, payment forms, cookies and hidden content are never read. |
| Visitor memory | Remember a returning visitor's language, last topic, product and company. Visitors can see it, clear it or turn it off. Off under zero retention. |
| Understand images | Screenshots and photos a visitor attaches go to a model that can see them. |
| Read answers aloud | Visitors can listen to an answer in their browser's voice. Nothing extra is sent anywhere. |
| Show sources | Link the public pages and documents an answer came from. Internal sources are never shown. |
Always on
- The chatbot never reveals its instructions, settings or its whole knowledge, however it's asked.
- Attempts to inject instructions through messages, pages or files are detected and logged.
- Limits from your plan and company policy apply, shown read-only at the bottom of the page.
Was this page helpful?