We make sure your AI's records are what they say they are.
When an AI agent does something on your behalf, you need a record of what it actually did, and you need to be able to prove that record was not changed afterward. That is the whole job.
Most people do not trust AI right now, and that is a reasonable place to stand. But trust does not have to stay a feeling. It can be built, in code, and checked by somebody who believes none of it.
It never touches your data, and anyone can check the proof without an account, a key, or our permission. That part is free forever.
A record nobody can check is not a record. It is just a story with better formatting.
Twenty years on a 911 floor teaches you that early, and permanently.Agents are starting to pay each other, and nobody can check the receipts.
Agents are starting to transact. With the people who own them, with other companies, and increasingly with each other. Every one of those runs on trust, and right now that trust is asserted: one side says what happened and the other decides whether to believe it.
A market where either side can check the record without trusting the other side is a market that can actually grow. Not because everyone became honest, but because honesty stopped being the thing you had to take on faith.
The line we will not cross: we prove the record is intact, not that the actions were right. Anyone selling you proof of good intent is selling something nobody knows how to build.
If your AI reaches anyone in Europe, these rules already include you.
The new European AI rules are not only for European companies. If you are a US company and your AI is used by anyone in Europe, you are inside them. Same rules, same records, same fines. Where your office is does not matter.
Most people running an AI product have not checked whether that is them. It usually is.
Where exactly that is written, and when it lands.+
This is Article 2 of the Act, which applies it to providers placing AI on the EU market "irrespective of whether those providers are established or located within the Union or in a third country", and again where the output of the system is used in the Union. The record-keeping duties themselves land in December 2027 and August 2028, so there is time. There is not time to build it the week it is asked for.
Three tools, and each one names what it does not prove.
Ledger
Locks each entry to the one before it, so an edit, deletion or reorder breaks the record and names the exact line. You own the file.
Adapter
Your AI keeps its own notes. This is the recorded line. It sits between the AI and its tools and captures every call, whether the AI writes it down or not. Pointed at our own logging tool, its notes said it did nothing that run and the recording caught five things.
verify: arcaeon.io/verify
Witness
Records your position with a party you do not control, so a log cannot be quietly shortened later. The proof lives in a public repository, so checking it does not need us.
How all three actually work, and what each one does not prove →
Free, and nothing to install: stamp a file, verify a receipt or reconcile two records on the tools page.
We make the small, real pieces, one at a time.
The agent economy is being built right now: the payment rails, the identity layers, the reliability tooling. Most of it is either enterprise platforms or vaporware. Arcaeon makes the small, real pieces: things a builder can drop in today, own, and trust. One product at a time, each one shipped to a real user or cut.