Search projects, hackathons, sponsors and tech

    ← All projects

    Agentry

    A Ledger-secured, Graph-powered autonomous purchase agent on Hedera Network

    Demo

    About

    A Ledger-secured, Graph-powered autonomous purchase agent. It pays for goods and services over Hedera's x402 rails, gates every payment behind a physical Ledger Nano approval, and reasons over its own on-chain purchase history — indexed by a subgraph — to suggest restocks and compare prices across vendors. The commerce-agent architecture draws on established patterns from prior Zepto402/Maid402-lineage builds, but Vault402 itself — the Ledger payment gate, the Hedera/Blocky402 settlement path, the purchase-log contract, and the subgraph — is built fresh for this event.

    How it's made

    Ledger's role is scoped entirely to protecting the key behind the Hedera payment. It is not involved in the Ethereum Sepolia write at all — that's a separate, ordinary-key transaction, since Ethereum Sepolia's only job here is being a chain The Graph can index. Decides — the agent identifies a purchase to make (restock or new order). Unlocks the key — the Hedera operator's ECDSA key is encrypted at rest via the Ledger Key Ring (ring encrypt). The agent never holds it in plaintext; right before settlement it calls ring decrypt, which requires the physical Nano to be connected and confirm before releasing the key into memory. Note this is a gate on the key, not a hardware-produced signature — Hedera isn't a wallet-cli-supported currency, so the actual ECDSA signing happens in software once the key is released, not inside the device's secure element. Settles — the agent uses that briefly-decrypted key to build and submit the payment through the Blocky402 facilitator, settling on Hedera testnet in HBAR. Logs — the agent calls PurchaseLog.recordPurchase(...) on Ethereum Sepolia, using a standard key from .env — no Ledger involvement on this leg. Indexes — a subgraph deployed to Ethereum Sepolia picks up the PurchaseLog event live. Reasons — the agent queries its own purchase history through the Subgraph MCP in natural language, computes consumption patterns, and surfaces restock suggestions and cross-vendor price comparisons.