Roasby lets AI assistants act on ad accounts that spend real money, so we design the product and the platform with safety and security in mind. This page summarizes the main measures we use today; they change over time as we improve them. They reduce risk but cannot remove it; see Limits. It is part of our Data Processing Addendum as the description of technical and organizational measures.
Safe by design
- Confirm before writing. Assistants can read freely and are designed to ask you to confirm before changing anything.
- Paused by default. What Roasby creates starts paused, so it should not spend until you turn it on.
- No delete tools. Roasby offers pause, but no tools to delete campaigns, ads or keywords.
- Guardrails. Budget limits and per-account change rate limits are there to catch runaway automation.
- Change log and revert. Changes made through Roasby are recorded with who made them and the values before and after, and most can be reverted in one click (spend that already happened cannot be undone).
Accounts and access
- Passwords are stored only as strong one-way hashes. Sign-in supports passkeys, two-factor authentication and Google sign-in, with rate limits against guessing.
- Workspaces are kept separate: queries for ad data are scoped to one workspace you belong to, your membership and role are checked on every request, and automated tests check this.
- AI assistants connect through OAuth with short-lived access tokens (1 hour), and you can disconnect them at any time.
- Our administrators must use two-factor authentication. Admin actions are recorded in an audit log.
Data protection
- Traffic is encrypted with TLS (HTTPS) and passes through Cloudflare's network protection.
- Ad platform OAuth tokens, two-factor secrets and other secrets are encrypted at rest.
- Records use random, non-sequential identifiers.
- Servers are hosted by Hetzner in Germany and Finland and run under a dedicated, least-privilege system user. Production access is limited to authorized personnel.
- Nightly backups are encrypted before they leave the server, stored off-site and checked for freshness, with automatic rotation.
- Personal data is kept only as long as described in the Privacy Policy, and old records are pruned automatically.
Software and monitoring
- Changes to Roasby go through automated checks before release (static analysis, tests, dependency vulnerability audits) and a security review.
- Errors are monitored with Sentry, configured not to send names or email addresses. Health checks alert our team about failures, full disks or stuck queues.
- Dependencies are kept up to date, and security fixes are applied promptly.
Incidents
If we discover a security incident that affects your personal data, we will contain it, investigate, and notify affected customers without undue delay (and, for customers under the DPA, within 72 hours of becoming aware of it), as well as authorities where the law requires.
Limits
No product or system is completely safe or secure. Software has bugs, ad platforms and AI assistants can fail or change how they work, and AI can misunderstand a request. Despite the measures on this page, a safeguard might not work in a particular case, an unwanted change might reach an ad account, or data might be exposed. Review the changes you confirm, watch your ad accounts and budgets, use two-factor authentication or passkeys, and tell us at [email protected] if something looks wrong. Our responsibilities are set out in the Terms of Service.
Reporting a vulnerability
Please email [email protected] with the details and steps to reproduce. Our security.txt has the same contact.
- We aim to acknowledge reports within 3 business days and to keep you informed until the issue is fixed.
- If you act in good faith, avoid privacy violations and service disruption, only test against your own accounts, and give us reasonable time to fix the issue before disclosing it, we will not take legal action against you for your research.
- Do not run denial-of-service tests, spam, social engineering or physical attacks, and do not access or change other users' data.
- We do not run a paid bug bounty yet, but we are glad to credit you.