Free tools

It reads three properties and explains them.

Spec version, manufacturer and firmware version — all three already published by Windows. From those it tells you whether the chip can produce attestation worth verifying, and whether SEVEN KG can check it against a vendor root we hold.

Download tpm-check.ps1 and run it:

powershell -ExecutionPolicy Bypass -File tpm-check.ps1

Some systems will not describe the TPM to an ordinary account. If it says it could not read the chip, run it again from a PowerShell started with Run as administrator — that is a permission problem, not an answer about your hardware, and the script says so rather than guessing.

It sends nothing.

There is no network call anywhere in it. It does not attest anything and produces no evidence. It is short on purpose, so that reading it before you run it is realistic rather than theoretical.

It will also tell you when the limitation is ours. An AMD firmware TPM is perfectly good hardware; we hold a root for it from a single source, which is not enough to anchor a production claim, so we refuse it and say why. You should expect a tool from a verification company to be straight with you about that.

You keep a signed result.

The agent asks our coordinator for a challenge, has your TPM sign a statement bound to it, and sends that back. What you get is a signed evidence bundle as a JSON file: the verdict, the values your hardware reported, and the full list of what the proof does not establish.

There is no public download yet. Ask us and we will send you the build.

The Windows build cannot do anything else.

It is compiled so that proving itself once and exiting is the only thing it can do. The work loop never starts, no job is ever claimed, and the machine registers with zero capacity. That is a property of the binary, not a setting you are trusted to leave alone.

Check it without trusting us.

The bundle carries an Ed25519 signature over its contents. Anyone can check it here by uploading the file, or offline with our own verifier. The certificates inside expire within minutes, so the recorded result and its timestamp are the durable part — nobody can recompute that answer later, ourselves included.

Not every chip is accepted.

We verify against vendor roots we have sourced and confirmed from an independent second source. Where we have only one source, we refuse the evidence rather than anchor a claim to a fingerprint we cannot corroborate. Every root we pin is published with its digest, including the gaps.

It is a brake, not a quota.

The counter is held in memory, so it resets whenever the coordinator restarts. Every deployment hands every address a fresh proof. We would rather say that than let the published number imply an accounting we do not actually do.

Two machines behind one router share one address, and each machine is also counted on its own.

Most homes and offices present a single address to the internet, so a second machine on the same connection will be refused until the cooldown clears. Each machine also has a cooldown of its own, so proving the same machine twice does not work from a different connection either. Both are deliberate, and the first is a real limitation of the free tier rather than a fault at your end. A paid attestation has no cooldown and can be made whenever you want, with the cost shown before you commit to it.

What is in it

Everything we worked out that quarter and can publish without naming anybody's device: Seal Check results, vendor root rotations, what the catalogue now covers, and what moved in the wider field — physical memory attacks, silicon that gained or lost a capability, a chip that turned out not to behave the way its own specification reads.

Every issue goes through the same publication review as everything else we put out: no per-device identifiers, ever, and every sentence read back as though by the owner of the device it describes.

Why it is free

Free Service.

Because we remember how hard this was to learn. Nearly every answer in here cost somebody an afternoon at a bench with a real machine in front of them, and most of them are written down nowhere else. Put a price on that, and the people who need it most — someone trying to work out what their own fleet can actually prove, before anybody has sold them anything — are precisely the people who never get to read it.