Monad Metropolis / agent commerce

AgentPay

A working agent that can buy its next breath.

A scripted story about an agent preserving its work and purchasing the next inference. Model usage and Monad settlement are historical evidence; the coding task is a separate teaching fixture.

RECORDED TESTNET TRACE

DEMO MODE / VERIFIED TESTNET REPLAY
NETWORK / EIP-155:10143

SCRIPTED STORY + HISTORICAL TESTNET PAYMENT. No new model call or payment is made. Capacity indicators are illustrative, not measured balances. The recorded model output did not generate this repair. Run the local acceptance tests ↗

WORKING / CHECKOUT RETRY TEST 2 OF 3

STORY RUN / KITE-07 / JOB #418VERIFIED PAYMENT / MONAD TESTNET

AUTONOMOUS MAINTAINER KITE-07 WORKING

INFERENCE FUELcapacity pressure detected

bounded refill available

ACTIVE JOB#418

Fix checkout retry loop 2 / 3 tests passing

$ npm test checkout-retry
FAIL retry preserves idempotency
agent: inspecting payment state...

Kite-07 is already working when the story begins.

The job is real to the agent: one failing test still blocks delivery.

SCENE 01 / WORK

KITE-07

checkout retry
job in progress

MARKET

aiplans.dev
+ toolkit.fun

POLICY WALLET

owner-signed
spend boundary

0.100000 ceiling

API PROVIDER

GLM-4.7-Flash
machine route

MONAD

Permit2 / upto
actual settlement

0.000001 USDC

JOB #418 2 / 3 tests passing

model metadataquote / digestreservationprovider callchain evidence

aiplans.dev/en/api-pricing OPEN LIVE ↗

aiplans.dev API Pricing Comparison page showing model, channel, provider, and China-access filters — LIVE SITE CAPTURE / SEP 2026 **333 models · 476 API channels · 21 providers**

GLM-4.7-Flash channels + merchant routePRICE SOURCE ↗

Zhipu direct API GLM-4.7-Flash / aiplans.dev

CNY 0 / 0input / output per 1M

ACCOUNT CHECKOUT

OpenRouter API GLM-4.7-Flash / aiplans.dev

USD .06 / .40input / output per 1M

API CREDIT

Toolkit metered API GLM-4.7-Flash / testnet demo rate

2 / 8 ATOMICinput / output per 1M

AGENTPAY READY

KITE-07 SELECTED Machine checkout + instant fulfillment policy match

REAL AIPLANS PAGE + ILLUSTRATIVE AGENT DECISION OVERLAY

toolkit.fun/app#api OPEN LIVE ↗

toolkit.fun application with the API and MCP distribution section selected — LIVE SITE CAPTURE / SEP 2026 **API / MCP distribution surface**

toolkit.fun / AGENT API SETUP Provision machine capacity

CONFIGURING

endpoint/api/agentpay/v1/models/zhipu/GLM-4.7-Flash/invoke

modelzhipu/GLM-4.7-Flash

authorizationPermit2 / upto

testnet rate2 in / 8 out atomic USDC per 1M

Permit2 ceiling100,000 atomic / 0.100000 USDC

credentialdelegated / no copied key

BUY agentpay.buy({
  body: { prompt: "resume job #418" },
  inputTokens: 1024,
  outputCap: 256
})

DISCOVERY POLICY PERMIT2 API READY

200 OK Model result attached to job #418 usage: 17 input / 256 output / 273 total

REAL TOOLKIT PAGE + ILLUSTRATIVE PROVISIONING OVERLAY

Evidence stream

Every visible event maps to a protocol transition.

00:00.00Goal receivedcheckout retry patch in progress

00:02.60Capacity threshold crossedbelow finish threshold / checkpoint saved

00:05.60Machine offer selectedreal prices compared / AgentPay route selected

00:11.10Budget reservedceiling bound to request digest

00:14.90Permit2 authorized0.100000 USDC Permit2 ceiling

00:20.10Capacity restored17 input + 256 output = 273 tokens

00:23.90Settlement matched1 atomic USDC paid on Monad

00:28.10Job resumedcheckpoint restored / tests pass

Spend ledger

Permit2 ceiling0.100000 USDC

actual settlement0.000001 USDC

unused under ceiling0.099999 USDC

proof idwaiting-for-receipt

Verified trace economics

Public market prices, bounded request, provider usage, and chain settlement from the same trace.

PROOF / MATCHED

Market / Zhipu direct CNY 0 input / 0 output per 1M

Market / OpenRouter USD 0.06 input / 0.40 output per 1M

Toolkit testnet demo rate 2 input / 8 output atomic USDC per 1M

Bound request 1,024 input estimate / 256 output cap

Toolkit offer limits 128,000 max input / 4,096 max output

Actual provider usage 17 input + 256 output = 273 tokens

INTEGER METER ceil((17 x 2 + 256 x 8) / 1,000,000) = 1 atomic Preflight worst case is also 1 atomic for the bound request.

MONAD SETTLEMENT 1 atomic = 0.000001 USDC rate toolkit-demo-2026-09 / Permit2 ceiling 0.100000 / unused 0.099999 USDC

Proof explorer

A receipt is useful only when a human can recover the same evidence.

transaction0xc1b583...263ed2

provider usage17 in / 256 out / 273 total

model resultGLM-4.7-Flash delivered

statusawaiting settlement

source transaction public proof