Aegis: Principal-Protected Hedging for the Agent Era
Turns "I'm worried about ___" into a principal-protected hedge in two clicks.
Tell the agent your concern: inflation, gold volatility, oil shocks, anything you can name. A TEE-verified AI on 0G Compute picks the asset that hedges it. You deposit. If it goes up, you win. If it goes down, you get every dollar back. Always.
Crypto promised self-custody and got casinos. AI promised assistants and got opaque chatbots. Aegis fuses them: an agent that talks to you in plain English, makes a financial recommendation you can cryptographically verify, and settles a hedge on-chain where your principal is mathematically protected by the contract. No counterparty trust. No "trust me bro" inference. No way to lose your deposit.
Anyone with $50 and a worry. Retirement savers hedging inflation, freelancers hedging oil, anyone tired of "high yield" products that are really just leverage in a costume.
Every autonomous agent shipping today (Claude, Cursor, GPT, custom) that needs a callable financial primitive it can offer to its user without taking custody, without burning audit trail, without exposing the user's wallet.
This is the first time an autonomous agent can make a financial recommendation that's provably honest. The model lives inside a TEE on 0G Compute. The reasoning is signed. The agreement is sealed onto immutable storage. The settlement is on-chain. The full trail, from the sentence you typed to the dollars in the vault, is auditable forever, by anyone, without asking us.
Browser (Wagmi + RainbowKit) ◀──── 0G Aristotle (16661) ──── AegisVault.sol + AUSDC.sol
│ ▲
│ REST │ approve, createShield
▼ │
Aegis API (Express)
│
├──▶ 0G Compute (TeeML inference, signature-verified) ─▶ NIM fallback (silent)
└──▶ 0G Storage (Indexer.upload, rootHash recorded on-chain)0G Compute (TEE) runs the inference inside an enclave. Response signature, provider address, and model are persisted on every shield, so the agent can't lie about what it recommended.
0G Storage holds the immutable agreement doc. The rootHash is sealed into the on-chain Shield struct, so the doc and the contract are inseparable.
0G Chain (Aristotle mainnet, 16661) hosts AegisVault and AUSDC. Shields live on-chain as a Shield[] mapping per user, with a mathematical clamp guaranteeing principal return at maturity.
OpenClaw skill manifest + /.well-known/aegis-skills.json let any agent autodiscover Aegis and call it on a user's behalf with a wallet-derived session key. The user signs once. The agent never touches the user's wallet.
AegisVault: 0x60403dd3CC683F65Db6dEb8597051aDc80506C3F
AUSDC: 0xA3CD4843Fc8f2Af53fa4786b16F70c90BfecD2F2
First mainnet shield: chainscan.0g.ai/tx/0x5657bf84…85bc130
Open the app, type a worry, sign three popups, and your principal-protected position is live on 0G Chain, with the AI's recommendation, the agreement doc, and the settlement all linked by cryptographic proof.
The future of agentic finance is one that can't lose your money. Aegis is what that looks like.
Over the hackathon window, Aegis went from concept to a fully working principal-protected hedging product live on 0G mainnet. Every layer of the stack got built, wired together, and shipped end-to-end.
Smart contracts deployed to 0G Aristotle mainnet (chain 16661). AegisVault.sol and AUSDC.sol are live and verified. AegisVault holds the Shield[] mapping per user and enforces the mathematical clamp that guarantees principal return at maturity, so the "every dollar back if wrong" promise is a property of the contract, not a policy. AUSDC is the settlement token wired into the vault's approve and createShield flow.
TEE-verified AI inference on 0G Compute. The recommendation model runs inside a trusted execution enclave on 0G Compute using TeeML. Every inference returns a signed response, and the response signature, provider address, and model identifier are persisted on every shield. A silent NIM fallback was added so the user experience never breaks, but the verified path is the default. The result: the agent cannot lie about what it recommended, and anyone can audit the recommendation after the fact.
Immutable agreement docs on 0G Storage. Each shield generates an agreement document that gets uploaded through the 0G Storage Indexer. The returned rootHash is sealed into the on-chain Shield struct at creation time, binding the human-readable agreement to the contract state permanently. The doc and the position are inseparable.
Full-stack app shipped. A browser frontend built on Wagmi and RainbowKit connects directly to 0G Aristotle for the on-chain calls, and an Express API layer orchestrates the 0G Compute inference and 0G Storage uploads. The user flow is two clicks and three wallet signatures from typed worry to live position.
Agent-callable from day one. The OpenClaw skill manifest and a /.well-known/aegis-skills.json endpoint were implemented so any autonomous agent (Claude, Cursor, GPT, or custom) can autodiscover Aegis and call it on a user's behalf using a wallet-derived session key. The user signs once. The agent never touches the user's wallet. This makes Aegis a callable financial primitive for the agent ecosystem, not just a standalone dApp.
First mainnet shield created. The full pipeline was proven end-to-end with a real principal-protected position on 0G mainnet: typed worry, TEE inference, signed recommendation, agreement upload, rootHash sealing, and on-chain shield creation. Transaction: chainscan.0g.ai/tx/0x5657bf84…85bc130.
Deployed addresses:
AegisVault: 0x60403dd3CC683F65Db6dEb8597051aDc80506C3F
AUSDC: 0xA3CD4843Fc8f2Af53fa4786b16F70c90BfecD2F2
By the end of the hackathon, Aegis was not a prototype or a testnet demo. It was a live mainnet product where a user could type a concern in plain English and walk away with a cryptographically verifiable, principal-protected hedge sitting on 0G Chain.
Aegis is bootstrapped. Every line of code, every contract deployment, every TEE inference call, and every storage upload that went into shipping this product was funded out of pocket by the team. There is no outside capital behind Aegis today, no venture round closed or in progress, no token sale, no pre-sale, no private allocation, and no SAFE on the cap table. What exists on 0G mainnet right now exists because the team chose to build it without waiting for permission or a check.
That choice was deliberate. Principal-protected hedging is a category where trust is the entire product, and we wanted the first version of Aegis to be something a user could verify end-to-end without having to ask who paid for it or what incentives were sitting behind the recommendation engine. The TEE attestation, the signed inference, the on-chain agreement hash, and the contract-enforced principal clamp all exist precisely so that the honesty of the system does not depend on the honesty, or the funding, of the people who built it. Bootstrapping the build kept that property clean.
The hackathon scope was funded entirely by the team's own time and resources. That covered smart contract development and mainnet deployment of AegisVault and AUSDC, the 0G Compute inference integration with TeeML signature verification, the 0G Storage pipeline for immutable agreement documents, the Wagmi and RainbowKit frontend, the Express API layer, and the OpenClaw skill manifest that makes Aegis callable by any autonomous agent. No grants were drawn down to ship the mainnet launch, and no third-party infrastructure credits subsidized the deployment.
Looking forward, Aegis is open to conversations with strategic partners, ecosystem funds, and 0G-aligned capital that can accelerate distribution, deepen liquidity in the vault, and expand the catalog of hedgeable assets and concerns the agent can underwrite against. The team is particularly interested in partners who care about the agent-callable financial primitive thesis, where autonomous agents need composable, custody-free, audit-friendly products to offer their users, because that is the surface area Aegis was designed to serve. Any future raise would be sized to extend that surface, not to validate the core product, which is already live and working on mainnet.
Until then, Aegis runs on conviction and shipped code. The contracts are deployed, the first shield is settled, and the roadmap from here does not depend on a financing event to keep moving.