The verifiable crisis-response guardian for DeFi & RWA. Risk scoring and exit decisions run on-chain in Arbitrum Stylus — execution is bounded, non-custodial, and reproducibly verifiable.
Aegis is a non-custodial guardian that protects DeFi and tokenized real-world-asset positions from exploits and depegs, by moving the decision of whether to exit onto the chain itself. Here is the problem we kept coming back to. When a protocol gets hacked or an asset depegs, the people who get out in the first few minutes keep their money, and everyone else absorbs the cascade.
More than three billion dollars was stolen in crypto in 2025, and the attacks that matter most happen at machine speed, usually while you are asleep. A whole market has grown up to solve this with bots that watch the chain and automatically pull your funds the moment a threat appears, and they genuinely work.
The catch is that every one of them makes the actual decision inside a closed backend running on someone else's servers. You cannot see why it would sell your portfolio, you cannot reproduce its reasoning, and if its key is ever compromised it can drain you.
You are trusting a black box with your money. Aegis takes a different path. We put the entire risk-scoring brain on the chain as an Arbitrum Stylus contract written in Rust, so every weight, every threshold, and the full algorithm are public and can be reproduced by anyone. You can run cargo stylus verify yourself and confirm that the contract live on Arbitrum is a reproducible build of the exact source in our repository.
The part that actually moves money is a separate, deliberately simple Solidity contract called AegisVault. It calls the Stylus engine on the hot path of the transaction, and it can only ever act inside hard limits the user sets in advance. It is triggered by a single keeper, it fires only when the engine returns an exit verdict, it can never move more than a capped amount, it can only convert into an asset the user chose, and it can only route through an adapter the user approved.
Even if the keeper key were stolen, the worst it could do is the one narrow thing the user already authorized. We think of this as binding the agent instead of trusting the keys. There is a specific reason this lives on Arbitrum. The scoring is genuinely heavy on computation. It walks over every incoming signal, aggregates them per attack class, runs a popcount over a bitmask that tracks up to thirty two distinct sources so it can tell when two independent detectors corroborate the same threat, and applies fixed point time decay so stale alerts fade out on their own.
That is exactly the kind of work Arbitrum tells builders to push into Stylus, where computation is priced ten to a hundred times cheaper than on the EVM. Running verifiable risk scoring on the chain is only affordable because of Stylus. And the seam at the center of the whole design, a Solidity contract calling a Rust contract directly on a transaction that moves real money, is the piece most teams will not even attempt. We did not simply assert the parity, either.
The Rust engine, a Solidity port of the same algorithm, and a TypeScript mirror all produce identical verdicts for identical inputs, so both the gas saving and the correctness are things anyone can reproduce. Sitting on top of the contracts is an autonomous agent. It gathers threat signals from Forta and from a trustless oracle deviation reading that is computed on the chain inside the Stylus engine itself, asks the engine for a verdict, and lets the bounded keeper fire the exit when the threat corroborates. It sends a live Telegram alert the instant it acts, with a link straight to the transaction on Arbiscan, and for borderline threats it asks the user to confirm before anything moves at all.
The fair way to see where we fit is that the players already in this space have proven the market is real and valuable. What none of them can offer, because of the way they are built, is a decision you can actually verify and an executor you never have to trust. That is precisely what Aegis adds, and it is only possible because the heavy lifting runs cheaply and transparently on Stylus.
All of this is live and verified on Arbitrum Sepolia right now. You can try the app at https://aegissafe.vercel.app, the Stylus risk engine is deployed and verified at 0xdC832Fac3C211E1148D00624c992299B2d954f17, the AegisVault executor is at 0x8Ac8baCc02F6a605f89D01bCa6d4A500fc525e7E, and the full source is at https://github.com/big14way/aegis.