Aroha Protocol Docs

Your agent is a brain plus a harness. Aroha stands at the harness's three narrow waists — tool calls, delegation, and irreversible actions. Start here to see where everything fits.

A firewall for your AI agent's tools. Wrap any MCP server with action limits, human approval gates, and an audit log — zero code, one config change.

Stage 1 (localhost, zero infra) → Stage 2 (deploy + auth) → Stage 3 (Hub registration). Each stage is independently shippable.

Send live requests to pre-built sandbox agents — no setup required. Great for exploring the protocol before writing a single line.

Full Aroha 1.0 spec: five-layer architecture, DID format, envelope schema, saga pattern, settlement, reputation.

Community → domain → org-signed: how Aroha cryptographically binds agents to their publishers via Ed25519 keypair signing.

Intent → Cart → Payment chain. Ed25519-signed spend limits that flow through every agent hop — with monotonic narrowing and ledger enforcement.

Delegate research, email, scheduling, and any capability with cryptographic action limits, compute budgets, human gates, and TaskReceipt audit trails.

Multi-hop delegation networks: A → B → (C, D) with verified mandate chains, auto-attenuating ctx.delegate(), depth enforcement, and receipt trees.

Add one JSON block to your MCP client config — gives any AI assistant access to every Hub agent with auto-issued mandates and 5 discovery tools.

Aroha authorizes; it doesn't move money. The two-layer model, the settlement interface, the Stripe & AP2 adapters, KYC binding, and what a real payment integration requires — with honest status markers.

Keep an agent, MCP server, or skill off public discovery and control who may call it. Registry visibility and access grants, aroha-sig signed callers with an allow-list, and why hiding a record is not the same as restricting calls.

pip install aroha · asyncio-native · LangChain, CrewAI, AutoGen bridges. The fastest path for ML engineers.

Full protocol depth — mandate chain, SagaEngine, settlement, registry. 20+ packages for production deployments.

Hermes Agent, ZeroClaw, OpenClaw, Composio/TrustClaw, LangChain, AutoGen, MCP, Google A2A.

aroha.routing — ModelBudgetMandate, ModelSession, cascade controller, Thompson bandit learner, token tracking.

Tier 1 Minimal (L0–L1) · Tier 2 Standard (L0–L3) · Tier 3 Full (L0–L5). Self-certification instructions.

@aroha-sdk/audit — append-only hash-chained log with tamper detection and JSON Lines SIEM export.

Capability manifests, verifiable credentials, saga logs, DID documents — what's personal data and what's not.

EscrowSettlement, money transmitter risk, safe settlement patterns, jurisdiction notes (US/EU/UK/SG).

Three registry tiers: Local (free) · HTTP/fiat · EVM/staking. ETH volatility mitigation strategies.

Planned for Q3 2026 — zero external deps, stdlib ed25519, SagaEngine, SpendingMandate.

Planned for Q3 2026 — Java 21 records, Bouncy Castle crypto, Spring AI bridge, Maven Central.

One-click managed agent hosting from Studio — deploy without writing any infrastructure config.

Permitted uses, fork policy, Aroha-conformant badge, prohibited uses. MIT License with clear brand boundaries.

How to contribute to Aroha Protocol — conformance test submissions and feedback welcome.