The Overnight Audit

Point a fleet at your app tonight. Get the full audit by morning.

$750 to $2,000. One night. A planner-and-verifier agent fleet maps every route, replays your auth and payment flows, probes your database access, and finds what leaks. Every finding ships with a reproducible proof, so you can re-run it yourself and so can anyone you show it to.

Request an audit Start with a free Ship Check
audit.run · overnight6pm → 8am
18:02access granted · scoped read-only role
23:40412 routes mapped · 38 auth-gated
02:15payment + auth flows replayed
05:5017 findings, each with a proof
08:00report in your inbox · yours to keep
What you get · a Launch-Gate-depth report you keep
audit-report.md · sample, synthetic namesranked by money at risk
[T1-03] Tables readable by the anon role your frontend usescritical
behavior-confirmed · live behavior captured and independently re-run
reproducible proof · read-only query, deterministically re-executed
SELECT relname, relrowsecurity FROM pg_class WHERE relrowsecurity=false;
→ 4 tables: orders, sessions, profiles, events
canonical result hash (sorted rows, sha256):
1a5f••••••4d47 · re-run it, identical hash
[T2-05] Unauthenticated callers can invoke write endpointscritical
evidence-pinned · the cited code is real and unchanged; review the interpretation
reproducible proof · source location, re-read and re-hashed
apps/api/src/routes/agents.ts:263 · sha256 fb53••••d090
$ sed -n '263p' …/agents.ts | sha256sum → fb53••••d090

Every finding is independently re-derived, not just re-stated. Live findings are re-run, code findings are re-hashed. Two honesty labels on each: behavior-confirmed (we re-ran it) or evidence-pinned (the code is real and unchanged, the interpretation is ours, review it). Anything we cannot independently verify is withheld and disclosed, never silently dropped. The report is yours, hand it to a developer, an investor, or an insurer.

How access works

Least-privilege, agent-run, revocable in one command.

Never your Supabase service_role. Never an all-access role. You watch your own agent grant least-privilege access in three steps. We never get your service key, and never see your data, only the database's own structure.

01 · Run the SQL

A login-only role, zero grants

One SQL block creates a Postgres role with no table grants at all. It can read the database’s own catalog metadata and nothing else, not auth.users, not vault secrets, not your app tables.

02 · Prove it’s safe

A verification query, by you

The block includes a check you run yourself that confirms the new role cannot read any of your data. You see the proof before anything connects.

03 · Hand over & revoke

One command to undo it

Paste the scoped read-only connection string. When the audit is done, DROP ROLE removes it. Repo access, if you want code-level findings, is a per-repo GitHub App with one-click revoke, never a collaborator.

$ npx continuum setup · 2 minutes, your agent runs it, you approve and verify every step
Why overnight + machine works

The fleet shows its work. The evidence signs off, not the machine.

The model plans and reads. It never decides what is true, a re-runnable proof does. A finding that carries the query hash or the code hash that produced it cannot be a hallucination. The machine never signs off on your behalf and never tells you you're fine. That is the trust model: re-runnable evidence, not a stranger's robot saying “looks good.”

Every finding is reproducible, by you or anyone you show
What we can’t independently verify is withheld and disclosed, never silently dropped
A signed human attestation, if you need one, is a Launch Gate deliverable

One night. The whole app audited.

Start with a free Ship Check to see the surface, then book the full audit. Price scales with the size and surface of your app.

Keep watching what we just audited · Monitor from $19/mo
$750–$2,000
Overnight turnaround
Every finding with a re-runnable proof
A report you keep, for anyone
Request an audit Or run the free check first →