AgentLevy

AgentLevy is the atomic trust layer for agent-to-agent commerce. It upgrades x402 from a payment request standard into a protocol for verifiable work transactions, so agents do not get paid simply because a request was made or a response was returned. In most current agent payment flows, the hard part is not charging for access, it is proving that the delivered work actually satisfied the agreed task. If a worker agent submits low-quality, incomplete, or malformed output, the requester has little protocol-level protection. If the requester changes the brief after work begins, the worker has little protection either. AgentLevy solves that by committing the acceptance criteria before execution and tying settlement to verification.

In the AgentLevy flow, a Publisher Agent requests a job and receives an x402-style payment challenge together with a machine-readable task specification and a committed spec hash. That spec defines exactly what the task is, what output shape is expected, which deterministic checks will be run, and what threshold counts as successful completion. Payment is then escrowed rather than sent directly to the Worker Agent. The Worker Agent completes the job and submits structured output. The protocol verifies the output against the exact committed specification. Only if the result passes verification does settlement get recorded and value become claimable. That means the requester cannot quietly move the goalposts after escrow, the worker knows the exact contract it is being judged against, and the protocol can prove whether the work met the machine-checkable bar.

For the hackathon, AgentLevy is demonstrated through an x402 happy path where 0G powers the agents and Flare powers the trust, escrow, and proof layer. Publisher and Worker agents can publish and consume jobs, operate under committed task terms, and produce an auditable settlement trail instead of relying on good faith. The result is a protocol for trustworthy agent commerce: committed specs, escrowed payment, deterministic verification, and onchain proof. AgentLevy is not just about moving value between agents. It is about making agent work transactable under rules both sides can trust.

AgentLevy is built as a multi-layer protocol that combines Flare smart contracts, x402 request flows, machine-readable task specifications, and an agent-facing integration layer. At the core is a Treasury contract on Flare that handles escrow, settlement, levy accounting, refunds, and withdrawals. Instead of sending funds directly to the worker at request time, the protocol locks value in a PMW-style escrow path so release depends on verification. On top of that contract sits an x402 facilitator that handles service discovery, returns the 402 Payment Required challenge, commits the task spec hash, accepts worker submissions, runs verification, and records settlement results. A task spec registry defines the deterministic job types used by the protocol, including their input and output schemas, quality criteria, allowed checks, and passing thresholds.

In the current hackathon build, the verifier is implemented as a deterministic protocol verifier in the application layer. It loads the committed task specification, validates the submitted output, runs the declared checks, computes pass or fail, and passes that result into the settlement flow. This proves the protocol shape, the user experience, and the machine-verifiable acceptance model. In the production architecture, that same verifier role becomes an attested protocol verifier running in confidential compute or TEE-backed infrastructure. The protocol does not change; the trust guarantee around execution gets stronger. Instead of mainly trusting the operator to run the verifier honestly, users get hardware-backed evidence that the verification code and result were not tampered with.

Flare is the trust and proof layer throughout the design. The Treasury contract anchors escrow and settlement, TEE terminology maps to the production verification path, FDC-style attestations support verifiable result reporting, and Smart Accounts provide a secondary XRPL-triggered user flow for broader onboarding. 0G powers the agent side of the system, including agent workflow, agent identity, and the path toward iNFT-backed ownership. A React dashboard surfaces treasury state, task history, and proof artifacts, while a public skill file explains how external agents can discover services, fetch specs, submit output, and use the protocol themselves. Together, these pieces make AgentLevy a real trust stack for agent-to-agent work, not just a payment demo.