Open-source financial execution firewall for AI agents.
Railguard is an open-source financial execution firewall for AI agents. It applies policy before signing, binds authorization to recipient, asset, amount, and network, tracks uncertain broadcasts, verifies settlement, and emits tamper-evident evidence.
On Arbitrum Sepolia we deployed execution-hook contracts and verified a separate 0.01 USDC external-wallet flow through on-chain SETTLED evidence. The public Failure Lab demonstrates six payment-failure classes (client-side simulation). v0.1 alpha / testnet reference.
Judge path (no login): Demo https://youtu.be/L-Gss08bzR0 · Public proof https://railguard-site.vercel.app/proof/arbitrum-sepolia · Failure Lab https://railguard-site.vercel.app/attack · Site https://railguard-site.vercel.app/ · GitHub https://github.com/prasanthkuna/railguard-gateway
On-chain: hook 0x756829c3ab0eB02b22fe4D7C9E35252A9738965E · settlement tx 0x243ec1e8eb6a992a85a035af11edd5ff70d78e3b84e7eb6c78ab5badc409608d
Optional operator replay (WorkOS sign-in): https://prebroadcast.vercel.app/login