Financial controlsfor AI agents.
When an agent moves money, Axorum records or refuses it at the moment it happens — checked against two laws at once: the money must balance, and the action must be permitted. Every verdict comes with its reason.
Cash received in full, but only one month is earned this period; the rest is carried as deferred revenue, so the quarter isn't overstated.
Everyone else is building the rails that let agents move money. Axorum sits underneath them, on each participant’s own books: it does not move money, hold funds, or settle anything.
Money is starting to move at machine speed.
Every commit has to balance — twice.
One atomic step, two conservation laws. The money must balance, and the authority must balance: the action was permitted, and every obligation it touches is accounted for. It posts only if both hold.
The double-entry is constructed and checked before anything settles.
Permitted, obligated, or forbidden — with the reason and the delegation chain on record.
Not a yes or a no. One of five outcomes.
Every commit resolves to a plain-English verdict a controller can read — the formal reading stays behind an engineer toggle.
| Glyph | Outcome | Reading | State |
|---|---|---|---|
| Permitted | Allowed and posted, no open duties. | Posted | |
| Refused | Nothing crossed the books; the reason is plain. | Refused | |
| Override | A prohibition overridden by a named approver — on record. | On record | |
| Duty open | Posted, and it opened an obligation to discharge. | Open | |
| Duty discharged | Posted, and it closed an open obligation. | Discharged |
Every map of agentic commerce has a hole in the middle.
Rails, issuers, wallets, identity, and fraud tooling exist. The layer that decides whether an agent should be allowed to spend does not.
Don’t take our word for it. Watch it refuse a payment.
Two ways in, both free and both in your browser: run it and try to break it, or read the short course that builds the idea one journal entry at a time. Technical evaluators, start with the demo: it runs the real engine, not a canned recording.
The live demo, in your browser
Watch agents pay each other, then break the system on purpose: pull the plug on a machine, cut the connection between them. The books stay balanced, refused payments stay refused, and nothing posts that should not have.
How a ledger learns to say no
A short scrolling explainer written for finance readers. It starts with one journal entry and ends with a refusal, one step at a time, with no code and no jargon.
Start with the control questions you cannot answer yet.
Read either assessment before you talk to anyone, including us. Each one is a list of questions your close already depends on and no team can answer today for an agent. No form, no call, no pitch.
Built for platforms putting agents on the rails.
Design partners set the control layer their category standardizes on — and ship trusted transactional agents ahead of everyone else.
The future of AI-native is secure agentic commerce.
Bring it to your customers first.