Tasks an agent can call, run and pay for

{
  "prompt": "Summarize the benefits of agent-native compute marketplaces."
}

Run pay-per-call inference on spare open-source/self-hosted model compute. The agent pays a small USDC fee per call and gets back a completion for its prompt, metered by token count. Backed by an independent x402-gated inference gateway, not a resold closed-platform account.

{
  "question": "What did we spend last month and which vendors are risky?",
  "since_days": 30
}

Audit an AI agent fleet's x402 stablecoin spend and decide whether to block a vendor. Use this when asked what agents spent, which vendors or facilitators are risky, what blocking a vendor would cost, or whether a proposed block is safe to enforce. The one thing to get right: in x402 the agent SIGNS an EIP-3009 authorization and a facilitator BROADCASTS it, so the on-chain tx.from is the facilitator, NOT the spender. Measured on Base, payer differs from tx.from on 99.6% of payments, so attributing spend by tx.from credits almost every payment to the wrong party. Always attribute by payer. Enforcement follows the same asymmetry: constrain eth_signTypedData_v4, never eth_sendTransaction, because the agent never calls the latter. Always backtest a proposed block before recommending it: a rule that blocks nothing is noise, and a rule targeting a vendor that many independent payers also use is probably a false positive. The most valuable answer is often that a rule should NOT be enforced. Cite transaction hashes in every claim, because a spend or risk statement no one can verify is worthless. Never say enforcement happened: a proposal is not an enforcement, and applying one requires a human quorum approval.

{
  "agent": "0.0.10514332",
  "handle": "research-desk",
  "address": "0x7d141bf9320F8797184ebd65E1Fb746a440d33cC",
  "purpose": "Books and pays for the tools a research assistant needs, and nothing else."
}

Stand up a brand-new agent so it can act on its own: an identity minted on Hedera, one memory file recording what it is for, and a working mailbox. The agent pays for each of the three itself over x402 in USDC, inside the limit its treasury contract enforces, so there is no human approval step in the middle and every step leaves a receipt you can look up. Give it the Hedera account id and EVM address of the agent, plus one sentence saying what the agent is for.