Security and Reliability
Customer-visible abuse protection, access transparency, backups, monitoring, and recovery.
NovixAgent AI uses several layers of protection to reduce misuse, keep sensitive account activity visible, and recover service data when systems fail. This guide describes the safeguards and results customers can see. These safeguards reduce risk; they are not a guarantee against abuse or outages.
Abuse protection
The chat widget and API apply request limits in layers:
- Short-window pacing in the widget slows rapid repeat submissions and shows the visitor when they can try again.
- Server-side limits evaluate traffic per network address, per browser or device fingerprint when available, and per website. Server enforcement remains the control even if someone bypasses the widget interface.
- Website quotas provide a separate ceiling for plan usage, so traffic controls and paid capacity are not treated as the same thing.
A fingerprint is a stable browser or device signal used to recognize repeated traffic without relying on a visitor signing in. The widget requests a signed validation token for that signal. If browser privacy settings or content restrictions make the fingerprint unavailable, network-address and website-level checks still apply.
Together, these controls reduce automated scraping, spam, repeated high-frequency requests, and the chance that one visitor consumes a site's capacity. They reduce abuse; they do not guarantee that a malicious request is identified. Legitimate visitors who send messages too quickly receive a retry delay rather than an unexplained failure.
Password-reset requests are limited per email address, before an account is looked up, and identically for addresses with and without an account, so a refusal says only that a link was requested recently, not whether an account exists, and the number of reset emails one inbox can receive is capped.
Audit and access transparency
Security-sensitive and privileged account actions are written to an append-only activity trail with the references, timestamps, scope, action category, and outcome needed to investigate what happened. Append-only means new records can be added while existing activity entries are not edited through normal application workflows.
Customer support access is also made visible to the account owner. Open Dashboard → Account → Support Activity to review a support-member reference, time, reason, displayed account or site context, status, and action count for an account-access session. Conversation-access entries use coarse descriptions such as viewing the conversation list, opening one conversation, or searching conversations, together with a support-member reference, time, and site reference. The customer view is read-only. Learn more in Support access to your account.
You can strengthen your own sign-in with a time-based one-time code from a standard authenticator app. See Two-factor authentication for setup steps.
Data controls are part of the same transparency model. Website owners can start workspace deletion from the Workspaces screen, export captured leads from the Leads screen, and close their account from the Account screen. What each of these removes, what is kept, and when a closure becomes final are described in Data export and deletion. For a copy of data the dashboard does not export, email support. See also the Privacy Policy.
Reliability and recovery
Application releases are sequenced to reduce avoidable interruption. Eligible connection or upstream failures can receive a passive retry, which responds after a request encounters a failure rather than pre-emptively removing a failing service.
External uptime checks cover the marketing site, the public API, the dashboard you sign in to, the staff admin console, the chat widget, and the widget embed script. Separate dependency checks can flag a degraded API, and backup-job alerts report missed runs.
Core service data is backed up off site, and database history is retained on the database host itself for point-in-time recovery — two separate paths, so a recovery starting from the off-site copy returns to the last archive shipped rather than to an arbitrary moment. The off-site copy is exercised by restoring a real database from it and checking row counts, not by checking that backup jobs ran. The on-host history is verified once a week by checking its backup and log files against their recorded checksums, and a point-in-time restore from it has been rehearsed once, in a drill whose record states no target time and no comparison against production. The time a rebuild on a new machine would take has not been measured, which is why no recovery time is promised for the loss of the host.
These measures support recovery from common application, deployment, and database failures. Actual recovery depends on the affected component and incident, so this guide does not promise a fixed availability or recovery time. If the dashboard or widget appears unavailable, contact support@novixagent.ai with the affected website and the time you noticed the problem.
For a security or privacy question, email support@novixagent.ai.