Security by design
We design Datanito systems around least privilege, explicit authorization, secure secret storage, encrypted transport, input validation, isolation boundaries, auditability, and safe failure behavior.
Agent permissions
Tool-using agents create a different security boundary than ordinary chat. We aim to scope credentials to the minimum required capability, separate read and write permissions where practical, and introduce approvals for high-impact actions.
Application security
We use secure coding practices including prepared database queries, CSRF protections for state-changing web requests, output escaping, anti-abuse controls, security headers, and dependency maintenance. Sensitive credentials should be provided through environment configuration rather than committed source code.
Infrastructure and data
Production architecture may include segmented services, encrypted connections, access logging, backups, monitoring, and provider-specific security controls. Access to production systems and data should be limited to personnel and services that need it.
AI-specific security
We consider prompt injection, unsafe tool calls, data exfiltration, untrusted retrieved content, model manipulation, excessive agency, and secrets exposure as product-security concerns. Controls may include tool allowlists, context separation, sanitization, approvals, policy checks, output validation, and evaluation.
Reporting a vulnerability
If you believe you found a security vulnerability, contact [email protected]. Please avoid accessing data that is not yours, disrupting service, or publicly disclosing an unresolved issue before we have had a reasonable opportunity to investigate.