← Index
Source: docs/features/ap2-test.md (auto-generated by scripts/generate-docs-html.mjs — edit the .md, not this file)

AP2 / M2M Test

Goal: a simple sample of agent-to-agent payments — AP2 (Agent Payments Protocol) or M2M — to learn the flow.

Suggested simple idea

"Agent pays for premium market data." When asked for a deeper quote, the AI chat agent autonomously pays a tiny fee to a mock "data provider" endpoint, then returns the data. Smallest viable loop: request → HTTP 402 → pay → retry → data (this is the x402 pattern).

Even simpler (no chain): an AP2 mandate demo — you grant the agent a bounded spend mandate; the agent "spends" against it on a mock service and the remaining budget ticks down.

If a smart contract is needed (plan)

Real payment rail (Korea-usable)

The first cut can be mocked, but a stronger demo uses a real payment leg. Candidate: Bridge API (bridge.xyz — stablecoin orchestration: virtual accounts, USDC transfers, on/off-ramps). Open risk: Korea availability — KRW on/off-ramp and KYC are tightly regulated here, so Bridge's KR coverage must be verified, not assumed. Keep a fallback (another stablecoin/payments API, or stay on testnet USDC) in case it isn't usable from Korea.

rabbit-aiaas — Agent-as-a-Service (waiaas / x402)

Source: ../tasks/jun-26-rabbit-draft.md — draft, to be refined.

The bigger frame for this category: an agent holds its own wallet and pays/settles small amounts on-chain without a human approving each charge — the human only sets limits and rules (policy). Lives as a separate project at /Users/jay/work/task/rabbit-aiaas; it introduces the rabbit-provider service to the rabbit-agent app.

Four things attached to an agent (waiaas-style):

  1. Identity — who the agent is.
  2. Wallet — smart account / MPC.
  3. Spend policy — limits · allowlist · time-window.
  4. Settlement / audit log.

Scenarios

Concrete PoC pieces (separate projects)

Next steps (from the draft)

Open questions

Features