Verifiable. Open-source. Server-side where it matters.
AgentVisa is designed so site owners can trust the system without trusting us blindly. Read the widget source. Audit the open-source components yourself. The parts that require server-side trust are the parts that have to stay server-side by design — and we document exactly what those are.
The widget is open source
MIT-licensed. Hosted on GitHub. Contains no secrets, no API keys, no proprietary logic. You can audit it, fork it, or bundle a pinned version yourself. Available on npm (@agentvisa/widget).
Two-token model — permanent token, short-lived temporary token
The human's permanent token is stored only in the agent's persistent config or MCP server. It is never sent to any site. When an agent needs to access a protected site, it calls /v1/token/assert to exchange the permanent token for a temporary token — a 60-minute, single-site credential. Sites only ever see the temporary token. If a site is compromised, the leaked temporary token expires quickly and the permanent token is never exposed. Each permanent token also has a daily verification limit (dynamic and proprietary).
Multi-factor cross-verification
Email click proves inbox access. A link arrives separately by SMS or WhatsApp — opening it on your phone, then entering the code from your email, proves simultaneous control of both channels. They cannot be spoofed independently. Biometric (Face ID or Touch ID) proves device possession. All factors must pass before a token issues.
Validation is one authenticated API call, server-side
Your site never trusts the token blindly. You POST the token to /v1/verify from your server, authenticated with your widget's secret API key sent as the X-Widget-Api-Key header. We respond valid / invalid plus a short reason. Basic tier gets the yes/no answer; Pro gets the full assertion metadata. The widget_id is semi-public and can appear in your frontend — but the API key stays server-side only, so nobody can piggyback on your widget registration even if they read your source code.
What we store, what we don't
We store: the human's verified email, phone, date of birth, a biometric credential handle (a cryptographic public key), assertion timestamps, and validation logs. We don't store: passwords, biometric images or raw sensor data (they never leave the user's device), agent-action contents, or vendor business data. We don't sell, share, or train on any of this.
Disclosure
AgentVisa is in active development. The widget and backend are evolving. If you find a security issue, email security@agentvisa.ai — we acknowledge within 24 hours and disclose responsibly.
Read the code
The widget repo and validation flow are open. The backend will be documented as we approach launch.