Orbital-Sentinel
Autonomous AI arbitrage agent built on Chainlink CRE - 8 cross-workflow monitors for LINK DeFi ecosystem, with LAA (LINK AI Arbitrage) - a workflow that detects LINK arb opportunities on Curve using AI analysis and executes on them.
What it is
Orbital Sentinel is an autonomous AI agent platform that monitors the entire stake.link DeFi ecosystem using 8 Chainlink CRE workflows. The flagship LINK AI Arbitrage (LAA) workflow is live on the CRE mainnet DON, running 7x/day autonomously.
The problem: DeFi arbitrage bots make isolated decisions. They see one pool and one price. But stLINK/LINK arbitrage on Curve depends on Priority Pool queue depth, Morpho lending utilization, CCIP bridge status, oracle prices, and governance activity — data spread across 15+ contracts and external systems. Missing any signal means leaving money on the table or executing into a trap.
How it works: 8 CRE workflows read live Ethereum mainnet contracts via EVMClient — Curve StableSwap, Chainlink Data Feeds (LINK/USD, ETH/USD), staking pools, Morpho Blue markets, CCIP Router/OnRamp/TokenPool, and 50+ ERC20 balances. After all workflows complete, a composite intelligence layer feeds the full ecosystem context to GPT-5.3-Codex, which produces an ecosystem-aware recommendation that can override the isolated arbitrage signal. Every decision is backed by a keccak256 proof hash written to SentinelRegistry on Sepolia.
Real example: The isolated LAA signal says "EXECUTE" (17 bps premium, pool open). But the composite layer reads that Priority Pool queue is 365K LINK deep, stLINK/LINK depeg is at 95 bps, and Morpho utilization is at 89.65% — and overrides to "WAIT" with 0.94 confidence. That decision, with all cross-workflow metrics encoded, is verifiable on-chain.
Architecture: Phase 1 runs 8 CRE workflows (LAA on mainnet DON, 7 via local simulate.
Phase 1.5 runs composite AI analysis across all workflow outputs.
Phase 2writes keccak256 proof hashes to SentinelRegistry on Sepolia.
A Next.js dashboard surfaces workflow status, CRE signals, and on-chain proof history with Etherscan links.
How it Works
- CRE TypeScript SDK (@chainlink/cre-sdk ^1.0.9): all 8 workflows use Runner, handler, CronCapability, EVMClient, HTTPClient, getNetwork, encodeCallMsg, consensusIdenticalAggregation
- EVMClient.callContract(): reads 15+ Ethereum mainnet contracts — Curve StableSwap, Priority Pool, Arb Vault, Chainlink staking pools, reward vaults, Morpho Blue markets, ERC4626 vaults, CCIP Router/OnRamp/TokenPool, Chainlink Data Feeds, and 50+ERC20 balances
- Chainlink Data Feeds: LINK/USD, ETH/USD, POL/USD via AggregatorV3 latestAnswer() on mainnet
- CCIP monitoring: reads CCIP Router getOnRamp(), OnRamp paused() state, and
LockReleaseTokenPool rate limiters across 3 destination chains (Arbitrum, Base, Polygon)
- HTTPClient + consensusIdenticalAggregation: deterministic off-chain fetches for AI analysis (Claude Haiku for treasury risk, GPT-5.3-Codex for arbitrage + composite
intelligence) and governance data (Snapshot GraphQL)
SentinelRegistry.sol (Solidity 0.8.19, Foundry): minimal on-chain proof registry with Ownable2Step, duplicate prevention, input validation. Audited with 31 tests (17 unit + 7 fuzz + 7 deep audit), 80K fuzz iterations
- Composite Intelligence: cross-workflow AI synthesis reads all workflow snapshots, builds ecosystem context, and calls GPT-5.3-Codex for a recommendation that can override isolated signals - viem: on-chain interaction for recordHealth calls and HealthRecorded event log collection
- Next.js 15 + Drizzle ORM: analytics dashboard surfacing sentinel records from PostgreSQL
- Bun runtime: all workflow projects use Bun for fast dependency resolution and execution
Links
Created by
- Avi