Reference

AI Payment Infrastructure Glossary

The vocabulary of agent payments — the open standards, the wallet and policy components, the settlement layer, and the use cases that motivate it. Plain definitions, no hype, with links to deeper writing where it exists.

QBT Labs builds tooling and integrations around these emerging standards. Our market making and trading workflows are the live operating vertical that proves the infrastructure works under real volume. See the AI Agent Payment Infrastructure overview.

Standards & Protocols

The wire-format and handshake layers. These are the open standards that let agents and APIs exchange payment intent, authorization, and settlement.

x402

An open standard for programmatic, pay-per-request HTTP payments. The server returns HTTP 402 with pricing metadata; the client signs a payment authorization and retries. No accounts, no API keys, no subscriptions — just per-call settlement on-chain or via a facilitator.

AMP— Autonomous Machine Payments Protocol

A persistent payment-channel protocol for agent payments. The agent deposits funds once, then issues many signed off-chain headers, with periodic on-chain settlement. Optimized for high-volume agents that would otherwise generate thousands of micro-transactions; for 1,000 calls, x402 needs ~1,000 on-chain transactions, AMP needs about 3.

MPP— Multi-Payment Protocol

A higher-level routing layer for agent payments designed to interoperate with multiple rails (e.g. Tempo, Stripe). Sits above per-request protocols like x402 to handle cross-rail settlement, pricing aggregation, and provider selection.

HTTP 402— Payment Required status code

The HTTP status code reserved in the 1996 spec for "future use" and unused for nearly thirty years. Now repurposed by x402 as the trigger that initiates a programmatic payment handshake between an agent and an API.

MCP— Model Context Protocol

An open protocol for connecting LLMs and AI agents to external tools and data through a uniform tool-calling interface. QBT Labs ships OpenMM — an MCP server that exposes market making and trading workflows to agents — as the operating-vertical proof point that infrastructure for agent commerce works under load.

Wallets & Spend Controls

Giving an agent a wallet is the easy part. Production agent payments live or die on what happens before the signature.

Agent wallet

A keypair (and supporting infrastructure) an autonomous agent uses to sign payments. Distinct from a human wallet because it operates without UI confirmation. Production agent wallets enforce policy at the signer, isolate keys in a separate process, and emit receipts for every signed action.

Policy engine

The component that decides whether an agent is allowed to pay before any signature is produced. Enforces per-transaction caps, daily limits, address allowlists, chain restrictions, and time-of-day rules. Without a policy engine, "give the agent a wallet" is a blank cheque.

Spend limits / budgets

Hard ceilings on what the agent can sign for, expressed as caps per transaction, per recipient, per day, or per category. Enforced server-side at the signer, not client-side at the agent runtime. Spend limits are how product teams turn an unbounded LLM into a bounded financial actor.

Payment Mechanics

How money actually moves. The settlement layer — what gets recorded, when, and by whom.

Facilitator

In x402 flows, an optional third-party service that verifies payment authorizations and handles on-chain settlement on behalf of resource servers. Facilitators let API providers accept programmatic payments without running their own chain integrations or custody.

Settlement

Finalizing a payment by writing it to a chain — or recording it for later batch settlement. In x402, settlement is per request. In AMP, settlement is periodic: many off-chain micropayments collapse into one on-chain transaction. The right choice depends on payment volume, recipient count, and fee tolerance.

Use Cases & Concepts

What this infrastructure is for. The motivating workflows that make agent-native payments different from a human checkout flow.

Stablecoin micropayments

Payments denominated in stablecoins (USDC, USDT, and similar) at small per-call values — fractions of a cent up to a few dollars. The economic foundation of pay-per-request APIs and agent commerce; volatile native tokens make per-call pricing impractical for buyer and seller.

Machine-to-machine payments— M2M payments

Payments where neither side is a human at a checkout flow. An AI agent paying an API. A trading bot paying for market data. A workflow paying another workflow. Required when humans aren't in the loop and traditional billing primitives — invoices, subscriptions, manual approvals — don't apply.

API monetization

Charging for API access. Traditional model: subscriptions and API keys for human-managed accounts. Agent-native model: pay-per-request via x402, channels via AMP, with metering, receipts, and abuse controls. The agent-native model unlocks long-tail, low-volume, high-value access that subscription pricing can't reach.

Building on this stack?

QBT Labs ships open-source tooling around these standards and runs market making operations as a working proof point. If you're building agent payments, an MCP-exposed API, or an autonomous trading workflow, we can help.

AI Agent Payment Infrastructure Read the blog OpenMM on GitHub