Cookie su questo sito

Usiamo i cookie necessari al funzionamento del sito. Con il tuo consenso ricordiamo anche quale pagina o link ti ha portato qui la prima volta, per sapere da dove arrivano le attività che si registrano. Non usiamo cookie pubblicitari né di tracciamento di terze parti. Puoi cambiare la tua scelta in qualsiasi momento da “Impostazioni cookie” in fondo alla pagina. Leggi l'informativa sui cookie

Assistant Workshop
Accedi
Tema

Bozza

Questo testo non è ancora stato rivisto da un avvocato e i campi tra parentesi quadre sono ancora da compilare. Lo pubblichiamo perché tu possa leggere cosa offrirà il servizio; nella forma attuale non è ancora in vigore.

Security at Assistant Workshop

Questo testo non è ancora disponibile nella tua lingua; viene mostrato in Inglese.

Version 2026-10-05

This page explains how we protect the data of the businesses that use Assistant Workshop and of their customers. The threat model and the security policy published with the source code name the code and the tests behind each measure.

1. Where the data lives

  • The service, its database and the language model it uses by default run in the European Union.
  • Providers outside the European Economic Area are used only for the functions named in the list of sub-processors (section 8 of the data processing agreement, published at /dpa).
  • Backups are encrypted before they leave the hosting provider and are kept for 30 days (daily copies) and 12 months (monthly copies).

2. Who can see a business's data

  • Every request for a business's data checks that the person belongs to that business, and the database keeps businesses apart with row-level security.
  • Platform staff open a client's cabinet only with a stated reason, for at most 60 minutes and read-only unless the owner allows changes; the owner sees a banner and the access is recorded in the audit log.
  • Platform administrators sign in with an authenticator app, and sensitive actions ask for the second factor again.

3. Signing in

  • Sign-in uses one-time codes sent by SMS, WhatsApp, Telegram or e-mail; each code allows a few attempts and is stored only as a hash.
  • Owners can add an authenticator app, see the devices they are signed in on and sign any of them out.
  • Sessions end after a period without activity and after a maximum lifetime.

4. Encryption

  • All traffic goes over HTTPS, and browsers are told to use nothing else (HSTS).
  • Messenger tokens and calendar access are encrypted in the database with keys that are rotated.
  • Call recordings are stored encrypted, with a separate key for each business.

5. The assistant

  • What customers write is treated as untrusted input; attempts to make the assistant ignore its rules are detected and slowed down.
  • Before a reply goes out, a guard checks prices, times and other facts against the business's own information and hands the conversation to a person when it cannot confirm them.
  • Tools such as booking take the business from the conversation, never from what the model writes.

6. Records and control

  • Views of customer data, exports, erasures, team changes and administrator actions are recorded in an audit log the owner can read.
  • Owners export their business's data and erase a customer's data in the settings, and personal data is deleted after the retention period the owner chooses.

7. How the service is built

  • Every change is checked automatically: tests, static analysis and scans of dependencies and secrets (CodeQL, bandit, pip-audit, npm audit, gitleaks, Trivy).
  • Third-party build steps are pinned to exact versions.

8. Reporting a vulnerability

Please report a security issue privately, as /.well-known/security.txt describes. We acknowledge a report within 2 business days, and we tell the affected clients about a personal data breach within 48 hours of becoming aware of it.