Problem

Cross-border payment companies process billions in daily FX transactions. When exchange rates move sharply during settlement windows (T+0 to T+2), manual monitoring misses critical

moments — leaving no verifiable evidence for post-incident audits.

The gap: "AI makes decisions" and "AI decisions are auditable" are worlds apart. Most AI systems today produce outputs with no cryptographic trail.

Solution

FX Risk Agent is an autonomous AI agent that creates an immutable, verifiable audit trail of every decision on the 0G Network — not just "what the AI said," but "what the AI knew,

judged, and recommended at each moment."

How it works:

1. AI analyzes real-time FX data for 4 currency pairs (USD/CNY, EUR/USD, GBP/USD, USD/JPY)

2. Every decision — full reasoning + model metadata — permanently stored on 0G Storage

3. Risk alerts recorded on-chain via FXRiskOracleV2 with access control (only Agent INFT owner can submit)

4. Every session updates Agent INFT (ERC-7857 inspired) state → on-chain inferenceCount grows

5. Anyone can verify: on-chain rootHash → download full decision log from 0G Storage

Full 0G Network Integration (4/4 components)

Architecture Highlights

• Dual AI Backend: Doubao Seed 2.0 Pro (primary) + 0G Compute (Qwen 2.5 7B via TEE-capable broker)

• Access Control: mintAgent onlyOwner + submitAlert requires INFT ownership → prevents forged "official" AI alerts

• Reentrancy Protection: ReentrancyGuard on state-mutating functions

• Hybrid Storage: Summary on-chain (gas-efficient) + full decision log off-chain on 0G Storage (verifiable via rootHash)

Tech Stack

• AI: Doubao Seed 2.0 Pro + 0G Compute (Qwen 2.5 7B)

• Smart Contracts: Solidity 0.8.24, OpenZeppelin 5.6.1, Foundry

• Backend: TypeScript, Node.js, ethers.js v6

• 0G SDKs: @0gfoundation/0g-ts-sdk, @0glabs/0g-serving-broker

• Network: 0G Galileo Testnet (Chain ID: 16602)

Live Contracts (Galileo Testnet)

• FXRiskOracleV2: 0x2abde2687923ffb9a5be4c6df3aac68a4f0a93ca

• FXRiskAgentINFT: 0xAA540f42f0d20588f183E3B92B3b617991fa22D1 (Token #0 minted)

• FXRiskOracle V1: 0x12030bc39dd18E2e8e4F10e685b7B7E639F0925A (historical audit)

Live Demo

• 📊 Dashboard: https://smallironman666.github.io/fx-risk-agent/

• 📹 Demo Video: https://youtu.be/j2eaoJN18a8

• 🔗 GitHub: https://github.com/smallironman666/fx-risk-agent

Engineering Quality

• ✅ 28 Foundry tests passing (access control, reentrancy, DoS protection)

• ✅ 7 Architecture Decision Records (ADRs) documenting design rationale

About

Built by @0xSmallironman — 5 years cross-border payment infrastructure experience (FIX 4.4, SWIFT MT103, ISO 20022). "From SWIFT to Smart Contracts."

Full 0G Integration Live + Production-Grade Security (2026-04-14)

Since the Checkpoint submission, FX Risk Agent has completed all 4 0G infrastructure components and hardened the contract layer with production-grade security patterns.

🎯 4/4 0G Components — All Live on Galileo

🛡️ Security Architecture

• Access Control: FXRiskAgentINFT uses OpenZeppelin Ownable — only the official deployer can mint Agent identities, preventing forgery of "official" AI alerts

• Authenticated Alerts: submitAlert requires INFT.ownerOf(tokenId) == msg.sender — every on-chain alert is cryptographically bound to a legitimate Agent

• Reentrancy Protection: ReentrancyGuard on all state-mutating functions

• DoS Protection: Query pagination capped at MAX_QUERY_COUNT = 100

• Input Validation: Strict integer/range checks on all agent identifiers

📦 Live Contracts (Galileo Testnet)

• FXRiskOracleV2: 0x2abde2687923ffb9a5be4c6df3aac68a4f0a93ca

• FXRiskAgentINFT: 0xAA540f42f0d20588f183E3B92B3b617991fa22D1 (Token #0 minted)

• FXRiskOracle V1 (historical): 0x12030bc39dd18E2e8e4F10e685b7B7E639F0925A

End-to-end validation: one full agent session completed across all 4 components — V2 alert recorded, INFT state updated, storage rootHash verifiable on-chain.

📊 Engineering Metrics

• 28 Foundry tests covering access control, ownership transitions, DoS boundaries, and multi-backend scenarios

• 14 TypeScript integration tests for off-chain components

• 7 Architecture Decision Records documenting key design choices (dual-backend strategy, V1/V2 coexistence, TEE trade-offs, etc.)

• Solidity 0.8.24 + OpenZeppelin 5.6.1, built with Foundry

🌐 Global Accessibility

Dashboard migrated to GitHub Pages with auto-deploy workflow — globally accessible without regional DNS/CDN restrictions.

Live Dashboard: https://smallironman666.github.io/fx-risk-agent/

🔗 Key Links

• 🐙 GitHub: https://github.com/smallironman666/fx-risk-agent

• 📊 Dashboard: https://smallironman666.github.io/fx-risk-agent/

• 📹 Demo Video: https://youtu.be/j2eaoJN18a8

• 🔍 V2 Explorer: https://chainscan-galileo.0g.ai/address/0x2abde2687923ffb9a5be4c6df3aac68a4f0a93ca

🎯 Next Milestones (Before Final 5/16)

• Implement tokenURI for INFT visibility in wallets and explorers

• Record fresh demo showcasing the full verifiable decision flow

Not yet funded. Building as an independent project for the 0G APAC Hackathon.