The problem
Risk, priced and settled by agents.
On-chain protection markets normally depend on a trusted backend to feed in real-world facts, so users cannot verify why a policy paid out or not. NERVE turns external events — a BTC price threshold, rainfall, an outcome — into auditable on-chain state without trusting a hidden oracle.
How it works
The mechanism
A user buys a parametric policy in native STT, escrowed by the RiskDesk contract. The contract requests the resolving fact through a Somnia JSON-API agent (fetchUint) returned under a multi-validator median-of-5 consensus, plus an LLM inferString leg that tags a volatility tier without touching escrow. The callback writes the observed value on-chain, a permissionless keeper can resolve any expired policy, and the contract pays out and emits a receipt trail when the trigger condition is met.
Agent pipeline
Who does what
Scout
Watches real-world signals and flags the risk worth covering — the live path is BTC downside.
Underwriter
Quotes premium and payout with deterministic v1 pricing, then writes the policy on-chain.
Resolver
Fires the Somnia JSON-API agent request for the resolving data and awaits the callback.
Auditor
Writes a human-readable receipt: source, callback status, decision, payout, and tx hash.
Policy resolved on-chain — JSON median price ≤ threshold → status Paid, payout sent (resolution reqId 3737216). View RiskDesk on the Shannon explorer ↗
Screens
/desk
A live trading-desk shell: ticker tape, a four-stage agent pipeline that lights up as Scout→Auditor fire, a monospace activity tape, and a receipts drawer with the on-chain resolution.
Run it yourself