AgentTrust Receipts for the work your AI says it did
A service from Syrup Labs

The free AI health check

Your agent says it worked. We check.

Automations report success all the time. Some of it is not real. A sync that has no credentials still reports success. A dashboard shows green while it grades an empty index. A notification queue fills up and delivers nothing. You find out when a client asks.

AgentTrust audits what your agents actually did and hands you the evidence. In 48 hours, free, we find three things your dashboards are getting wrong.

No install. No access to your accounts required to start. Three findings or we say so plainly.

Findings, one estate, one morning 12 SEP 2026
Gateway status: OFFLINE
Gateway answered every request beside the probe. The health check called a method the gateway does not implement and rendered the error as an outage.
91 tasks waiting on you
The database says 18. Both numbers are right. Neither screen says what it counts.
Memory: 11 files, 614 KB
The next tab on the same screen, same store: 0 files, 0 chunks.
483 reports published
329 of them were referenced by no decision, ever. 63% of everything written.

Measured on a live 98-agent estate. Every line above is a real finding with a real observable behind it, and every one had been true for weeks before anyone looked.

How the health check runs

Three days, and you do almost none of it.

The order matters, so it is numbered. Each step depends on the one before it.

DAY 1

You show us what runs

A 20-minute call. Which automations run without you watching, and what each one is supposed to produce. No credentials, no installs.

DAY 2

We chase every claim to its observable

For each reported success we find the thing that would exist if it were true: the delivered message, the written file, the row in the table, the fetched page.

DAY 3

You get the findings and the receipts

A short report: what is real, what only claims to be, and the evidence for each. Yours to keep, and yours to forward to a client.

What gets checked

Five questions your dashboard is not asking.

These are arithmetic, not opinion. Fetch a URL. Read a status code. Ask the filesystem whether a path exists. There is no model deciding whether your work happened.

IDENTITY

Is anything actually running?

An agent in a dashboard can be a database row. We check whether a runtime exists behind the name.

DELIVERY

Did the message arrive?

A queue with 1,348 rows and zero deliveries reports success in most tools. We check the far end, not the send.

OUTPUT

Do the links resolve?

Every URL, file path and endpoint an agent handed you, fetched. A 404 is not a completed task.

CONSUMER

Is anyone reading it?

Reports that no decision ever referenced are work you paid for twice: once to produce, once to ignore.

HONESTY

Does the indicator mean what you think?

We record the question each status light actually asks, then compare it to the one you assume it asks. They differ more often than not.

EXIT CODE

And the one rule under all of it

An exit code is not proof. A job can run, find nothing to do, touch none of the things being tested, and exit clean.

Fit

This is worth nothing to some people. Honestly.

Worth it if

  • Work runs when you are not watching: overnight jobs, scheduled syncs, automated outreach
  • Someone else asks whether it ran. A client, a boss, an auditor
  • You read dashboards and do not read traces, and do not intend to start
  • You have already been embarrassed once by an automation that said it worked

Not worth it if

  • You watch every run yourself. Nothing can fail invisibly, so there is nothing to find
  • You are still deciding whether to use AI at all. Come back when something runs
  • You are an engineer who will write these checks yourself. You should. We will tell you how
  • Nobody outside your head ever asks for proof

After the free check

If you want it standing rather than one-time.

The health check costs nothing and ends with a report. If you want the checks running after that, this is the shape that exists.

Start here
Health Check
Free once

48 hours. Three findings in your own estate, with the evidence behind each. No obligation and no install.

First 30 days
$2,500 / 30 days

The checks run on what your automations claim, with the evidence behind every finding, and a report at the end of the month you can hand to a client.

After that
Monthly or stop

Continue month to month or stop. The monthly price after the first 30 days is not set yet, and it will be agreed before month two starts.

Questions

What people ask before the check.

What does the free health check find?

Three findings in your own estate, with the evidence behind each. It takes 48 hours, with no obligation and no install.

Who is this for?

The solo operator of an automation agency running 1 to 15 client accounts on unattended n8n, Make or Zapier workflows, who has been asked by a client whether something actually ran.

What happens after the free check?

If you want the checks running, the first 30 days are $2,500. After that you continue month to month or stop.

What does the monthly plan cost after the first 30 days?

That price is not set yet. It is agreed with you before month two starts.

What if everything checks out?

Then you will be told so plainly, and you will have lost nothing but one short call.

Find out what your dashboards have been telling you.

Tell us what runs unattended and what it is supposed to produce. Three days later you will know which of it is real. If everything checks out, we will say so and you will have lost nothing but one short call.

Or just reply with the one thing that has been nagging you. That is usually where the first finding is.