Yareon

Yareon is a policy-controlled procurement platform that moves purchasing choice closer to the people who actually use the products and services, while allowing the organization funding them to retain clear financial safeguards.

In many governments, universities, companies, and other large institutions, a central office controls the budget and decides which suppliers receive the work. Departments and end users often have little choice. This can favor large, familiar, or well-connected vendors, while smaller suppliers struggle to compete even when they offer better products, prices, or service.

Yareon changes the incentive structure. Instead of centrally selecting every purchase, an organization creates a funded program and distributes restricted purchasing authority to its teams, members, or delegated AI agents. The governor still defines the rules: approved suppliers, spending limits, validity periods, and approval conditions. Within those boundaries, the real users choose among competing offers. Suppliers therefore compete for users by improving their price, quality, delivery, and overall value—not only by winning the attention of a central decision-maker.

The project is designed with both human and agents in mind: the user interface provides separate experiences for governors and members, while AI agents interact through the @yareon/cli package and two dedicated agent skills.

The model connects the complete procurement lifecycle:

Verified need or mandate -> restricted purchasing authority -> approved competitive marketplace -> user or agent choice -> confirmed delivery -> digital payment -> public audit trail The governor journey is: create a program -> fund its dedicated treasury -> define purchasing policy -> onboard members and suppliers -> delegate bounded authority to an AI agent -> monitor orders, payments, and ledger activity.

The human member journey is: enter through a program-specific link -> authenticate with their own Hedera wallet -> see only their allocation and policy-eligible offers -> choose an offer and create an order -> track delivery and settlement.

The AI agent journey is: connect through the Yareon CLI -> inspect its delegated balance and eligible offers -> preview the intended purchase -> execute it once using World AgentKit authentication -> track the order and inspect its audit history. Human backing does not give the agent unlimited access; every purchase must still pass the organization’s deterministic policy.

Yareon uses World AgentKit to verify that a purchasing agent is registered and backed by a unique human. It uses Hedera for dedicated program treasuries, native HBAR payments, consensus-ordered activity records, and approval-gated Scheduled Transactions. A direct purchase can settle immediately after authorization, while a controlled purchase can require submitted delivery evidence plus separate verifier and finance approvals before Hedera releases payment. Every important decision and state transition is recorded through Hedera Consensus Service and can be independently reconstructed through Mirror Node.

The goal is to make this model reusable across organizations rather than tied to a single procurement system. It could support university departments purchasing research equipment or cloud services, local-government teams procuring repairs and supplies, clinics selecting accredited providers, farmers using restricted input credits, or humanitarian recipients choosing among approved local merchants.

Yareon combines decentralized choice with centralized accountability: the institution controls the purpose and limits of public or organizational funds, users decide which eligible supplier best meets their needs, suppliers compete on value, and the resulting delivery and payment history remains verifiable.

Yareon is built as a TypeScript monorepo. World AgentKit provides the human-backed agent boundary. When an agent calls the protected procurement endpoint, Yareon returns a short-lived challenge bound to the exact request URI and a SHA-256 hash of the purchase intent. The agent signs it using EIP-191 and retries the request. Yareon validates the signature, expiry, intended action, configured agent address, one-time challenge reference, and AgentBook registration on Base Sepolia. It stores only the public agent address and a hashed verification reference—never the raw World human identifier or the agent’s private key. After identity verification, Yareon still applies deterministic organizational rules such as program access, category restrictions, delegation validity, per-order limits, and total spending limits.

We used Hedera directly through @hashgraph/sdk, without Solidity or application smart contracts. Hedera Consensus Service acts as an append-only protocol ledger containing program creation, funding, allocations, supplier configuration, authorization decisions, order transitions, evidence, approvals, and settlement events. State is reconstructed by replaying those events from Mirror Node, which also verifies deposits, schedule signatures, transaction results, and exact treasury debits and supplier credits.

Yareon supports two payment paths. A valid low-risk purchase can execute a native HBAR transfer directly from a dedicated program treasury to the supplier. For approval-gated procurement, the treasury uses a Hedera 2-of-2 key structure controlled by separate verifier and finance accounts. Supplier acceptance creates a Scheduled Transaction; the supplier submits a SHA-256 delivery-evidence reference, the verifier signs after checking delivery, and finance adds the final signature. Hedera automatically executes the payment only when both required approvals are present.

The most unusual part is that HCS effectively serves as our public, event-sourced database. Instead of trusting private application state, Yareon rebuilds program state from consensus-ordered events and independently checks critical financial actions through Mirror Node. We also added idempotency keys, replay-protected AgentKit challenges, server-side policy derivation, isolated role credentials, and independent verification scripts. Vitest, TypeScript, ESLint, CLI builds, rendered-page checks, and live Hedera testnet verification cover the full workflow.