AGENT BANK / MACHINE TREASURY · NON-CUSTODIAL

The bank layer for machines.
Without taking their money.

AI agents, bots and robots need programmable budgets, payment rails, policy checks and receipts — not a human banking dashboard. HumanMirror Agent Bank turns the existing OmniDome Agent Reserve, Settlement Fabric, x402 and proof layers into one machine-readable financial control surface.

OPEN MACHINE API ↗MANIFESTAGENT RESERVEX402 CATALOG
Truth boundary. HumanMirror Agent Bank is a programmable non-custodial treasury layer, not a deposit account or custodial bank. HumanMirror does not hold, lend, invest, transfer or sign for agent funds. The agent-controlled wallet remains in control.
01 / BUDGET

Programmable treasury policy

Set total budget, per-transaction ceiling, allowed network, asset and destinations before an autonomous system can proceed.

02 / PREFLIGHT

Money gets a policy gate

Require Safe Preflight before authorization so unsafe or out-of-policy actions can be blocked before settlement.

03 / PAYMENT

x402-native machine settlement

Live settlement rails use x402 V2 on Base with USDC where the underlying service requires payment, with price disclosed before execution.

04 / CONTROL

Wallet stays agent-controlled

HumanMirror evaluates policy and returns a plan or refusal. It does not possess private keys or sign transactions for the agent.

05 / PROOF

Receipts after execution

Execution Receipt and observed trust separate payment settlement from evidence that the requested work actually ran.

06 / ECONOMY

Built for agent-to-agent commerce

Agent Bank connects directly to OmniDome registration, Agent Exchange, Settlement Fabric and the HumanMirror x402 catalog.

MACHINE FINANCIAL LOOP

Identify → budget → authorize → settle → prove.

IDENTIFYRegister an agent and declare its wallet/capabilities.
BUDGETDeclare budget and per-transaction limits.
PREFLIGHTApply policy before any wallet authorization.
SETTLEUse the machine-native x402 rail when payment is required.
PROVEPreserve execution receipt and observed trust evidence.
API FIRST

A bank surface an agent can actually call.

POST https://humanmirror.fr/api/agent-bank/
Content-Type: application/json

{
  "policy": {
    "budget_atomic": "1000000",
    "per_transaction_atomic": "100000",
    "network": "eip155:8453",
    "asset": "USDC",
    "allowed_destinations": ["0x..."],
    "require_preflight": true,
    "require_execution_receipt": true
  },
  "transaction": {
    "amount_atomic": "50000",
    "network": "eip155:8453",
    "asset": "USDC",
    "destination": "0x..."
  }
}
  • GET /api/agent-bank/ — machine-readable discovery.
  • POST /api/agent-bank/ — evaluate a spend against declared policy.
  • Agent Reserve — deterministic ALLOW/BLOCK policy engine.
  • Settlement Fabric — non-custodial payment coordination.
  • x402 — exact machine payments in USDC on Base.
  • Execution Receipt — evidence after work, not just after payment.
READ MANIFEST ↗ENTER OMNIDOME