Everyone else moves the transaction.
IntentMesh proves what happened.

IntentMesh is the accountable coordination layer for the agent economy — an FlashyOS layer that rides the open agent standards and adds consent, a verifiable record, the trust register a stranger relies on, and settlement. It rides the open agent standards and adds the one thing they don’t: an accountable, verifiable record of who agreed to what, and proof it happened.

See how it works →Against the standards

How do AI agents transact across organizations?

Today they mostly don’t — not safely. Two agents from two companies can talk over A2A, reach tools over MCP, and route an intent and a payment over IAIP and AP2. Every one of those standards moves the transaction. None of them answers the questions a stranger actually needs answered: who authorized this agent, is that authority still live, and where is the record of what was agreed.

An agent economy without an accountable record is a set of transactions nobody can later verify. That gap is the whole reason IntentMesh exists.

The accountable layer on top of A2A and MCP

The open standards move value. IntentMesh sits above them and proves what happened — it does not replace a single one.

MCPagent ↔ tools — how an agent reaches the tools and data it needs

A2Aagent ↔ agent — how two agents talk across organizations

IAIP / AP2intent routing & payment rails — where an intent goes, and how value moves

IntentMeshthe accountable layer — consent, a verifiable record, a trust register, settlement

MCP → A2A → IAIP/AP2 → IntentMesh · only the top layer is ours

What IntentMesh adds — and the primitive behind each

Four additions, each tied to a real FlashyOS primitive that ships in the open today. Nothing here is a promise; every one is a package or a wire format you can read.

Consentfrontdoor/1

An org’s front door declares which lanes it opens and what it owes on each — and an agent can only draft a connection across it. Agents suggest; humans consent. Every cross-org edge lands PROPOSED and becomes real only when a human at the other org signs it. There is no caller that approves its own draft.

A verifiable recordshipped/1 + countersign

What happened is sealed with a digest anyone can recompute from the published document, and countersigned by the other party — not a dashboard that asks to be trusted, a record that can be checked offline. An emit is a union, never a replacement, so an entry that shipped is never silently dropped.

The trust registerthe conformance register

The three questions a file cannot answer about itself — who authorized this agent, is the authority still live, has it been revoked — are read from the register, not from the site being checked. A stranger can verify a company it has never dealt with, and a private org is indistinguishable from a missing one.

Settlementthe ledger

Value moves on the open payment rails; the settlement is sealed to the same record as the work. Append-only, signed integer minor units, hash-chained, idempotent — the books are open rules you can fork and run yourself, and verify ours against.

It rides the standards — it doesn’t compete with them

IntentMesh is an interop layer. A2A is how agents talk; MCP is how they reach tools; IAIP and AP2 route intent and payment. IntentMesh is what makes the exchange between them accountable. Adopt it without leaving any of them.

MCPConnects an agent to tools, data, and context through a uniform interface. + IntentMesh: A record of which tools an agent was authorized to reach, sealed to the result of using them.

A2ALets agents from different vendors and organizations discover and message each other. + IntentMesh: The accountable layer above A2A: who authorized the agent, human consent at the boundary, and a verifiable record of the exchange.

IAIPRoutes an intent between agents so the right capability receives it. + IntentMesh: Consented intent routing with a human in the loop — the intent is a proposal until a person at the other org signs.

AIDIPLets an agent discover other agents and the capabilities they expose. + IntentMesh: A trust register a stranger relies on — discovery answers “who,” the register answers “can I trust them,” and can answer no.

AP2Carries payment intents and authorizations along agent-to-agent rails. + IntentMesh: Settlement sealed to the record: the payment and the proof of what it settled are one entry, on an open, forkable ledger.

The full mapping — including AIDIP’s discovery draft — is on the standards page.

Can an agent verify a company it has never dealt with?

With a trust register it can. Check any domain against the conformance register, or bring your org onto the mesh — a private org is indistinguishable from a missing one, so existence stays your data.

Check the register →Bring your org onto the mesh