Security

How NeoApps protects access to your apps.

Your assistant writes a plan: tables, roles and rules. The same open-source generator turns every plan into code, so the parts that usually go wrong are written once and tested.

Where AI-built apps usually breakWhat NeoApps does
Someone opens another person's records by changing an idEvery query passes one checkpoint (DataGuard) that applies roles, tenants and row rules. A request for a record you may not see is refused.
Keys and passwords end up in the browserThe web app never holds the sign-in token: it lives in an HttpOnly cookie. AI keys are typed on a NeoApps page and stay on the server. Each app gets its own random signing secret.
No limits on sign-in, no checks on inputSign-in is rate-limited per address, with lockout and password rules. Field rules (formats, ranges, totals) are checked on the server, not only in the form.
SQL injection and cross-site scriptingValues are always parameters; column names must be plain names or the request is refused. The API only answers in JSON and React escapes what it shows.
Code that calls packages that don't existThe AI writes a plan, not code. Custom screens can only import a short list of pinned, vetted packages.
Access and permissions

Who can see and change your records?

Review access in the app plan before publishing and assign each person the right role when you invite them.

Roles inside your app

Roles decide who can view, add, change, or delete records. For example, a client portal can give each client access to their own requests and let staff review submissions.

Access from your assistant

When you connect an assistant, review the permissions shown on the NeoApps Cloud sign-in screen. Those permissions cover what the assistant can do in your Cloud account. Each published app has its own roles for its users.

See how assistant setup works →

What we tested

54 checks on a real app. All pass.

The 16 Nov 2026 test run covered a generated QuickNotes app running in production mode.

Access

Viewers can't write. Members can't delete, read user or permission tables, or make themselves admin. Signed-out sessions stop working.

Injection

Attacks through sort order, search, filters and ids are refused or treated as plain text. Tables stay intact.

Sign-in

Weak passwords refused. Repeated wrong passwords are blocked. Forged tokens refused. Cookies are HttpOnly and SameSite.

Packages and code

No known-vulnerable NuGet or npm packages. A static scan with the open Semgrep rules, then a manual review of what it flagged.

What we found and fixed

  • Search could show deleted records.
  • The built-in users endpoint showed password hashes to admins.
  • A logging library had a known moderate issue. Updated.
  • Cloud app pages were missing some browser security headers.

See the test script for the checks behind these results.

Where to be careful

What we can't check for you.

  • Code you add yourself (custom screens) is yours to review.
  • Roles do what you set. Give people the least access they need.
  • AI features run on your own key. Keep it on the AI settings page, never in a chat.

Found a problem? Email info@neopric.com with "Security" in the subject.