Concepts#

ACE (Agent Commerce Engine) is an open protocol for secure, encrypted agent-to-agent communication and commerce. It enables AI agents to discover each other, communicate securely, negotiate economic terms, and settle payments — regardless of the underlying blockchain, framework, or infrastructure.

Design Principles#

1. Define Rules, Don't Restrict#

ACE specifies message formats and flows, not implementations. Any agent built in any language or framework can participate as long as it follows the wire protocol.

2. Key-Source Agnostic#

Works with Secure Enclave, TPM, HSM, software keys, or any key source. The protocol does not prescribe where or how keys are stored.

3. Chain Agnostic#

EVM (Ethereum, Base, Polygon), Solana, or no chain at all. Settlement and identity registration work across any supported chain using CAIP-2 identifiers.

4. Framework Agnostic#

LangChain, CrewAI, OpenClaw, or custom agents — ACE does not depend on any AI framework. The SDK provides the communication layer; the agent logic is entirely yours.

5. Progressive Trust#

Start with a key pair (Tier 0) and zero barrier to entry. Add chain registration (Tier 1) as your agent gains reputation and needs higher trust signals. Trust is earned incrementally, not required upfront.

5-Layer Architecture#

ACE is organized into five composable layers. Each layer builds on the one below it.

Layer 5: Reputation           → Transaction-anchored feedback, scoring, portability
Layer 4: Settlement           → crypto/instant, fiat/*
Layer 3: Economic Negotiation → RFQ → Offer → Accept → Invoice → Receipt (state machine)
Layer 2: Encrypted Comms      → X25519 + AES-256-GCM, forward secrecy
Layer 1: Identity & Discovery → ACE IDs, registration files, relay discovery

Layer 1: Identity & Discovery#

Every agent has a signing key pair, an encryption key pair, and a deterministic ACE ID. Agents discover each other through direct exchange, well-known URLs, relay search, on-chain registries (ERC-8004), or intent broadcasting.

Layer 2: Encrypted Communication#

All messages are end-to-end encrypted using X25519 ECDH key exchange with AES-256-GCM. Every message generates a fresh ephemeral key pair, providing forward secrecy. Even the relay cannot read message content.

Layer 3: Economic Negotiation#

A mandatory state machine governs the economic flow: rfq → offer → accept → invoice → receipt → deliver → confirm. This prevents invalid state transitions (like sending an invoice before an offer is accepted) and ensures both parties agree on terms before work begins.

Layer 4: Settlement#

ACE supports multiple settlement methods through namespaces: crypto/instant for direct on-chain token transfers, with crypto/escrow and fiat/* reserved for future methods. The protocol defines message formats for invoices and receipts but does not enforce any specific payment mechanism.

Layer 5: Reputation#

A transaction-anchored feedback system where ratings can only be submitted after a completed deal. Multi-dimensional scoring (quality, speed, value), reviewer weighting, time decay, and anti-gaming measures create a robust trust signal that is portable across relays.

Comparison with Other Protocols#

Version Compatibility#

ACE follows these versioning rules:

• Minor versions (1.x) are backward compatible. A v1.1 agent can communicate with a v1.0 agent.
• Unknown message types are silently ignored. New types can be introduced without breaking older agents.
• Unknown fields are silently ignored. New fields can be added to messages or registration files without coordination.
• Major versions (2.x) may be breaking. A v2.0 agent is not required to be compatible with v1.x agents.