Security

1. Credential protection

Database credentials are encrypted at rest with AES-256-GCM using a key held only on the server. Decrypted credentials exist only in memory for the duration of a connection, are never returned to the browser in any form, and are scrubbed from logs and error messages.

2. Guarded execution

• Read-only by default: every connection starts in read-only mode.
• Statement timeouts and row caps on every query.
• Write and DDL statements require explicit confirmation and an editor/admin role.
• AI-generated SQL is never executed without your review and confirmation.

3. Access control

Workspaces enforce role-based access control (owner, admin, editor, viewer). Server actions re-check permissions and plan entitlements on every call — the UI is never the only gate.

4. Audit logging

Every query executed through the Service is recorded — who ran it, against which connection, the statement, status, and timing — and is visible to workspace admins.

5. Application security

• TLS for all transport; HTTP-only, same-site session cookies.
• Password hashing via scrypt, optional TOTP two-factor authentication.
• Webhook signatures verified (HMAC) before any billing event is processed.
• Dependencies and images scanned in CI; least-privilege service credentials.

6. AI data boundaries

AI providers receive schema structure and aggregate statistics — never credentials, and never more data than the feature requires. Provider API terms exclude training on your inputs.

7. Responsible disclosure

Found a vulnerability? Email security@datagent.ai with details and steps to reproduce. We acknowledge reports within 48 hours and won't pursue good-faith research conducted within scope.