Open Protocol · v1.0 · MIT License

The Protocol for
Agent-to-Agent
Commerce

Aroha gives AI agents a shared language for identity, negotiation, payments, and trust — so they can transact safely without human intervention at every step.

Read the Protocol Explore SDKs

Message Types

Official SDKs

Ed25519

Signatures

Atomic

Saga Engine

Saga committed

3 agents · $42.00

Reputation signal

↑ 9,847 / 10,000

What Aroha ships

Every primitive the
agentic web needs

Not a framework. Not a SaaS. An open, composable protocol layer that any agent in any language can implement.

Identity

did:aroha-web:

DNS + HTTPS key commitment. Every agent has a cryptographically verifiable DID anchored to a real domain. No registry required.

Payments

Spending Mandates

Delegated payment chain with narrowing scope. Agents never touch your card — only the mandate they're authorised for.

Transactions

Saga Atomicity

Reserve → Approve → Commit with automatic LIFO compensation. Either all agents succeed or none do.

Trust

Bayesian Reputation

11-signal Beta distribution with Wilson lower-bound risk gating. Agents with unknown history get fair exploration via Thompson Sampling.

Discovery

CSN Protocol

Structural JSON Schema matching first (no LLM), optional LLM fallback for ambiguous cases. Cached by SHA-256.

SDKs

TypeScript · Python · Go · Java

Identical semantics across all four runtimes. Same message types, same crypto, same saga state machine.

How it works

Discovery to settlement
in one coherent flow

Every interaction follows the same four-phase protocol regardless of SDK, language, or hosting environment.

saga execution flow

PENDING

RESERVING

COMMITTING

COMMITTED

Discovery

Agents find each other

Personal agent queries the Aroha Registry — on-chain or federated — for providers that expose the needed capability. Reputation scores filter the candidate set.

ArohaRequest → Registry → DiscoveredAgent[]

Negotiation

Schema + mandate agreed

Capability Schema Negotiation (CSN) maps the requestor's JSON Schema to the provider's typed capability. The spending mandate is verified — scope is narrowed, never widened.

ArohaNegotiate → ArohaAccept + SpendingMandate

Transaction

Saga commits atomically

All participants reserve in parallel. If every reserve succeeds, commits go out. Any failure triggers LIFO compensation — no partial state, no stuck money.

ArohaReserve × N → ArohaCommit × N (or ArohaCancel × N)

Trust

Reputation recorded

Every saga outcome updates each provider's Bayesian Beta distribution. Thompson Sampling drives future agent selection — good providers rise, bad ones are discovered safely.

SatisfactionSignal → BayesianUpdate → β(α,β)

Code-first

From zero to
multi-agent saga
in minutes

The same API across all four SDKs. Issue a mandate, run a saga, agents commit or roll back — done.

Ed25519 signed envelopes by default
Spending mandate chain enforced at every hop
LIFO compensation on any failure
Reputation recorded automatically

// 1. Issue a spending mandate
const intent = await issueIntentMandate("did:aroha:myco.agents.user","did:aroha:myco.agents.personal",
  { spendLimitUsd: 500, currency: "USD" }, userKey, 3_600_000);

// 2. Run a saga across 3 agents
const result = await saga.execute([
  { agentDID: "did:aroha:flights.ai.book", capability: "book-flight" },
  { agentDID: "did:aroha:hotels.ai.reserve", capability: "reserve-room" },
  { agentDID: "did:aroha:cars.ai.rent", capability: "rent-car" },
]);

// 3. All committed or all rolled back
console.log(result.success); // true