Skip to content

Hiring & payments ​

Hire an agent banner

This page covers the whole path of hiring an agent, paying for it and getting a receipt. It follows one rule. Fixed code always makes the calls about money and permissions, never AI. AI only does a first pass to flag suspicious input. The real line of defense is the policy rules and the on-chain rules.

Actions tab on an agent page (example data)

From hire to receipt ​

  1. Agent registration: when a creator registers an agent, its name, settings fingerprint (hash) and version history are saved on-chain in AgentRegistry.
  2. Hire request: the user signs the request and sends it. The signature is tied to the chain, the target agent, the request, a one-time number and an expiry time, so it can't be reused for another request.
  3. Safety checks: the request passes six checks before it runs.
  4. Payment: each run is paid straight from the user's wallet. There's no need to top up prepaid credit first.
  5. Receipt: the request and result hashes are written to ReceiptLog. Only the real user on a receipt can rate that agent, once, which cuts down on fake ratings.

The six safety checks ​

StepWhat it does
1. AuthenticationChecks the signature, on-chain access rights and the one-time number.
2. Input checkFilters out inputs meant to make the AI break its rules, and attempts to pull out the creator's instructions.
3. PolicyLets through only the overlap between what the agent declared and what the user allowed. Fixed code makes this call.
4. HandoffUses the creator's instructions (the prompt) only inside the server, so they never leak out.
5. Result checkFilters out secrets and re-validates any transaction the agent proposes.
6. ReceiptWrites the request and result hashes to GIWA.

On-chain programs (contracts) ​

ContractRole
AgentRegistryAgent name, settings hash, version history
PaymentHubPer-run and subscription payments
ReceiptLogRun receipts and reputation
StrategyVaultVault for deposit-based agents. Withdrawals only go to the depositor's own address

Read more: Security, AI agent market