A human security audit for your AI-built app

You built it fast with AI. Before real users and real data arrive, have a security engineer review the app and its agent configuration — and hand you a findings report you can act on.

48-hour turnaround Delivered by a human No critical findings? It's free.

What you get

Fixed-scope review

A defined review of your app and its agent configuration — the same scope for every audit, so you know exactly what is covered before you commit.

Findings you can act on

Every item in the report carries a clear status — [PASS], [WARN], or [FAIL] — so you can see at a glance what held up, what needs attention, and what needs fixing now.

Delivered by a human

A security engineer reads your app and verifies every finding by hand before it reaches the report — no auto-generated noise to wade through.

48-hour turnaround

Your findings report is delivered within 48 hours of intake.

No critical findings? The audit is free.

If none of the critical checks comes back [FAIL], you pay nothing. You either get findings worth fixing, or you get the all-clear at no cost.

Three fixed tiers. No scope creep.

Audit + Report

$99 one-time

Know where you stand.

  • Fixed-scope review of your app and its agent configuration
  • Findings report with [PASS] / [WARN] / [FAIL] items
  • Delivered by a human within 48 hours
Most useful

Audit + Fix Guidance

$249 one-time

Know where you stand — and exactly what to change.

  • Everything in Audit + Report
  • Specific fix guidance for every [WARN] and [FAIL] finding
  • Findings prioritized by severity, so you fix the worst first

Audit + Remediation Call

$499 one-time

Walk through every finding and fix, live.

  • Everything in Audit + Fix Guidance
  • A live remediation call to work through the findings together
  • Direct answers on your stack, your fixes, your priorities

Every tier is a one-time purchase with the same fixed review scope — the higher tiers add fix guidance and a live call, not a bigger bill.

What we check

34 checks in 8 areas, the same for every audit. 19 of them are critical, or 16 when your app takes no payments: the audit is free when none of those comes back [FAIL].

  • Secrets and credentials

    5 checks, 3 critical

    Working keys, tokens and database passwords left in your code, in its history, or in the files your site sends to browsers.

  • Sign-in and permissions

    5 checks, 5 critical

    Whether every action that changes data needs a signed-in user, and whether one user can reach another user's records.

  • Database and file storage

    4 checks, 4 critical

    Whether your database, uploaded files or backups can be read or changed by anyone on the internet.

  • Input handling

    4 checks, 4 critical

    Whether text a visitor types can run as a database query, as a script in another visitor's browser, or as a command on your server.

  • AI and agent configuration

    5 checks

    Where your model keys live, what your agents and their tools are allowed to do, and whether untrusted text can steer them.

  • Dependencies

    3 checks

    Known critical flaws in the packages you depend on directly, and direct dependencies that look abandoned or impersonated.

  • Connections and browser protections

    5 checks

    Encrypted connections, browser protections, error messages that give away internals, and limits on repeated attempts.

  • Payments

    3 checks, 3 critical

    Whether prices are decided on your server, payment notices are checked before they change anything, and paid features are really gated. Skipped when your app takes no payments.

With access to your code we read the source. With only a link to the running app we test it from outside, and the report names each check that could not be reached that way. The audit is a review of your app as it stands on the day: it is not ongoing monitoring, not a certification and not a penetration test.

What the report looks like

An example of the layout, written for an imaginary booking app that takes no payments. It is not a customer's report.

Summary

[PASS]
24
[WARN]
5
[FAIL]
2
[N/A]
3

First finding

C1-1 · Bookings table can be read without signing in · [FAIL] · Critical

What we found
The bookings table answers any request that carries the public key shipped in the site's own code. No sign-in is needed.
Evidence
A request made with only the public key, and no signed-in user, returned rows from every account.
Impact
Anyone who opens the site can download every customer's name, email and booking history.
Fix guidance
Turn on row-level security for the bookings table and add a rule that returns a row only to the account that owns it.

Fix guidance comes with Audit + Fix Guidance and Audit + Remediation Call.

Request your audit

Send your details and we reply within one business day to arrange payment and access. The 48-hour clock starts at intake.

Which audit?

Questions first?

Email brad@deepsweep.ai — you will get an answer from the person who does the audits.