Security
AI-navi handles conversations between your site's visitors and an AI. These are the measures we take — and, for transparency, the ones we do not yet have.
Data protection
| Encryption in transit | All traffic uses HTTPS (TLS) with HSTS enabled. |
|---|
| Hosting location | The application and database run on a VPS in Japan (Xserver). |
|---|
| Tenant isolation | FAQ, navigation settings, and conversation logs are kept separate for each site (tenant). The widget does not work when embedded outside the domains a customer has registered. |
|---|
| Backups | The database is backed up daily (kept 14 days on the server) and also encrypted (AES-256) and transferred daily to a server at a different provider in Japan, where it is kept for 30 days. |
|---|
| Secrets | API keys are stored in an owner-readable environment file on the server, never in source code or public directories. Two-factor secrets are encrypted in the database. |
|---|
Access control
| Passwords | Stored as bcrypt hashes; never in plain text. Passwords must be at least 10 characters; common passwords and ones containing the site ID or email are rejected. |
|---|
| Login protection | Per-account and per-IP login attempt limits, email verification at sign-up (no login until verified), and notification emails for security-related actions. |
|---|
| Two-factor authentication | Customers can optionally enable TOTP two-factor authentication in the admin panel. Our own operator console also supports it. |
|---|
| Sessions | All sessions are invalidated when a password changes. Cookies are HttpOnly, Secure, and SameSite=Lax. |
|---|
| Operator access | Operator access to customer accounts is limited and recorded in an audit log. |
|---|
AI and personal data
| What is sent to the AI | To generate replies, visitors' questions and excerpts of site content are sent to Anthropic (Claude API). If whole-site learning (RAG) is used, passages of page text and the question text are also sent to a semantic-search API (Voyage AI) to find relevant pages (we have opted out of their use for model training). Under Anthropic's commercial API terms, this content is not used to train models. |
|---|
| Data minimization | No advertising IDs are collected and no identifiers are stored on visitors' devices. Unique-visitor counts use a hash that changes monthly, computed server-side. |
|---|
| Conversation log retention | Customers choose 3–36 months (default 12); older logs are deleted automatically. Bulk deletion and CSV export are available in the admin panel. |
|---|
| Safe outbound fetching | Features that read a customer's URL (site learning, navigation scan) reject internal-network addresses and limit redirects, size, and time. |
|---|
| Sub-processors | See our sub-processor list. |
|---|
Operations
| Monitoring | Server CPU and memory are monitored hourly with alerts; malware scans (ClamAV) run regularly. |
|---|
| Incident response | We have documented procedures for detection, containment, assessment, notification, and prevention. Events involving personal data are reported to affected customers without undue delay (generally within 48 hours). |
|---|
| Contracts | A Data Processing Agreement (DPA) template is available on request. |
|---|
| Operator | WeMot Co., Ltd. (Tokyo, Japan). Contact: info@ai-navi.site. |
|---|
What we do not have yet
So you can assess us accurately, we do not currently have or do the following.
- We do not currently hold third-party certifications such as ISO/IEC 27001 (ISMS), SOC 2, or PrivacyMark.
- We do not currently apply application-level encryption at rest to stored data (except certain secrets).
- A contractual uptime SLA is not offered on standard plans (available for Enterprise on request).
- We have not yet had a third-party penetration test performed.
If you need security-questionnaire answers or a DPA for your review, contact info@ai-navi.site. (Updated September 25, 2026)
