Yes.

The loss lands on the company; the question of what you were watching lands on you.

After August 2026, Article 14 of the EU AI Act requires human-on-the-loop oversight — a named supervisor able to interrupt the system at any moment.

+1Delaware law, since Caremark (Del. Ch. 1996), already puts directors on the hook for failing to implement and monitor a reporting system — not for the misconduct itself, for having no way to see it.

You do not need the agent stopped. You need to be able to prove you could have checked.

Your current software logs are debatable — a write happened; whether the agent stayed in its lane when it did is a separate question, and self-reported logs can't answer it. The hardware-verified Ship's Log is what we built instead — patent-pending (US 19/637,714) and already running.

Read the brief →License the floor — $20/agent-yr →

Would rather look at one than read about one? See what a receipt actually looks like — every commit we make mints one, and we did not remove the unflattering ones.

the log itself is free & MIT · the $20 is the seal that makes it evidence

we never guarantee the agent stays in its lane — only that the deviation is detected, placed, priced, and dispatched

how the ship's log is actually built — two moving parts

01 · FREE · MIT · RUNS ALONE

what it is · what it costs · what you get · why it's not the ask

Start with the free half.

thetacog-mcp is the open-source depository — the standardized format for logging what your agents actually did. MIT licensed, runs locally, nothing leaves your machine.
One line for whoever on your team owns the deployment: npx -y thetacog-mcp@latest attest-demo. It prints a signed local receipt in under a minute — no account, no key, no call to us. You do not need to run it yourself; you need to know it returns the same thing on their machine as on ours.
The code stays open so your own people can read every line of it rather than trust us, and so a receipt recomputes on an auditor's machine with nothing of ours in the room. That is what open buys. What it does not buy is the seal, and an unsealed log is still only your word for it.

THE 1+101 / 04

View the repo → github.com/wiber/thetacog-mcp

Making the log format open removes the entire adoption argument: nobody has to trust our server, ask our permission, or send us their data to start using it. You can spend your own engineering budget building whatever custom agent-management stack you want on top of it — the format underneath stays the same, which is what makes it checkable later.

The trap, if you want to call it that, is the next step. No matter how much you spend on your own internal AI oversight, you still can't hand a self-reported log to an underwriter and call it evidence. That's step two.

02 · $20 / AGENT-YEAR · THE CRYPTOGRAPHIC SEAL

why free isn't enough · what the seal does · who can check it · the price

The charge is the seal, not the software.

A log your underwriter cannot verify is not evidence. $20 per agent-year is the cryptographic stamp that seals it.
The seal makes the log recomputable by a stranger — a regulator, a reinsurer, the insured itself — offline, on their own machine. That's the whole difference between “trust me” and “check me.”
Price is fixed and published, not negotiated per deal. The term each license covers shrinks for later buyers, so early is the most coverage per dollar.
Complete on its own: one sealed, third-party-verifiable receipt for one agent, for one year. Nobody has to buy more than that to have something real.

THE 1+102 / 04

How the seal works → /iamfim

03 · WHAT THE UNDERWRITER DEMANDS

who wants it · who makes it · proof it's real, not a pitch

The receipt is the whole point.

The underwriter demands it. The deployer downloads the free repo to generate it. The seal is what turns it into evidence they can act on.
This isn't hypothetical — browse the receipts the same pipeline has already minted, in production, on this repository's own commits.
You don't need to buy anything to check this claim. The receipts are public, and every one recomputes to the same answer on a stranger's machine.

THE 1+103 / 04

Browse the signed receipts → /commit

04 · ONE DOOR, BOTH ENDS

the price · where it goes · why this is the first ask on the page

Swipe to certify.

$20 per agent-year. One license, one checkout — no separate sales call required.
This is the same door whether you arrived from a sticker, a blog post, or the book. Every path on this page ends at the same place the deck ends: /pricing — both doors, one checkout.
This is the first ask on the page. Everything above it — the repo, the explanation of the seal, the public receipts — was already yours for free.

THE 1+104 / 04

Get your seal — $20/agent-year → /iamfim

THE 1+1 · WHAT $20 BUYS1 / 4

If it’s debatable, it’s not insurable.

We sell the receipts

Still here?

Then you are the one who checks before signing, which is the only reason any of this works. Here is the stack: the ten-frame brief, the public adoption board with the unflattering numbers left in, the three hardening documents, and the advisory invoice.

What this is for

Directors’ and officers’ cover answers a question about what you were watching, not about what the machine did. A receipt does not settle that question and nobody can promise it will — but it is the kind of contemporaneous, third-party recomputable record an oversight defence is built out of, and the absence of one is the gap Caremark describes. Your broker decides what it is worth at renewal. We license the measurement and nothing else.

The stacked proof → /resources Or swipe · $20/agent-year

There is a physics underneath all of this, and none of the above depends on it. If you want the mechanism rather than the outcome, ask for it: /physics.

The builder’s door·Signed receipts·The licence